MSc thesis
(Thesis Code: DEC- 80433)
Determinants of Corruption of Village Chiefs:
Evidence from a Field Experiment in Rural Liberia
By
Mohammed Besher Degnet
(Development Economics Group)
Supervisor
Gonne Beekman (PhD)
(Development Economics Group)
Wageningen University and Research Centre
December, 2015
Acknowledgement
First of all, I would like to extend my deepest gratitude to my supervisor Gonne Beekman for
her constant, timely and invaluable feedback. I learned a lot from you about research and
scientific writing more than I did in the past 2 years. Dear Gonne, thank you for your support
and understanding given the later unforeseen developments during the process of writing the
thesis. Secondly, I would like to thank my study advisor Sudha Loman for her constant advice
and support during the course of my study. I am also grateful for the members and
contributors of Anne van den Ban Fonds (ABF) and particularly Mrs Frederique Six for
providing the full financial support for my study.
I would also like to thank my family, friends and classmates for their support and
encouragement. Last but not least, many thanks for Aziza. You are the best thing I have and
thank you for your patience, support and love. And, this is for you and for the would be new
member of the family.
Abstract
We investigated the micro-determinants of corruption of village chiefs in Liberia. We
hypothesised that corruption of chiefs is determined by their socio-economic characteristics
and of their villages. We used data collected from 136 villages and around 2600 households in
rural Liberia. We employed an objective measure of corruption by identifying the amount of
rice seeds missing after they are kept at the chiefs place. We estimated a tobit (censored)
regression model since our dependent variable is censored. We find that chiefs who were born
in the same village are corrupt as compared to their counter parts. We also find that chiefs
under a (perceived) corrupt district commissioner are corrupt as compared to their
counterparts. The results show the influence of leaders on their subordinates. All of the other
variables included in our study do not show statistically significant effect on corruption of the
chief.
List of Figures and Tables
Figure 1. (Macro-level) Determinants of Corruption................................................. 19
Table 1. Main Empirical Studies on Robust Determinants of Corruption................. 22
Table 2: Description of Variables................................................................................ 35
Table 3. Proposed Variables and their expected signs............................................... 36
Table 4.1 Summary statistics of variables................................................................... 44
Table 4.2.1 Determinants of Corruption: Main effects................................................. 49
Table 4.2.2 Determinants of Corruption: Interaction effects........................................ 51
Table of Contents
Acknowledgement....................................................................................................................................ii
Abstract ...................................................................................................................................................iii
List of Figures and Tables ...................................................................................................................... iv
Chapter One: Introduction................................................................................................................... 1
1.1. Background of the Study .......................................................................................................... 1
1.2. Statement of the Problem .......................................................................................................... 4
1.3. Objective of the Study and Research Questions ...................................................................... 7
1.4. Organization of the Thesis ........................................................................................................ 7
Chapter Two: Literature Review and Theoretical Framework ........................................................ 9
2.1.
Definition and Measures of Corruption .............................................................................. 9
2.1.1.
Definition of Corruption ............................................................................................... 9
2.1.2 Measures of Corruption .................................................................................................. 10
Theoretical Framework and Empirical Literature ............................................................. 12
2.2.
2.2.1.
The Incentives for Corruption and the Individual Decision Maker: An Economic
Perspective ....................................................................................................................................... 12
2.2.2.
A Sociological Perspective on the Drivers of Corruption: The Social Learning
Theory 16
2.3. The Opportunities for Corruption: A Brief Review of the Empirical Literature on the
Determinants of Corruption ........................................................................................................... 18
2.4.
A Brief Review of the Empirical Literature on the Consequences of Corruption ........ 24
2.5.
A Historical Overview of Leadership and Corruption in Liberia .................................. 27
Chapter Three: Data, Model specification and Estimation Strategy.............................................. 32
3.1. Data Source and Study Context .............................................................................................. 32
3.2. Model Specification and Description of Variables ................................................................ 32
3.3.
Hypotheses and Expected Signs ......................................................................................... 36
Chapter Four: Results and Discussions ................................................................................................ 44
4.1 Summary Statistics of Chief & Village Characteristics ................................................................. 44
4.2 Econometric results .................................................................................................................... 46
4.2.1 Determinants of Corruption: Main effects .......................................................................... 47
4.2.2 Determinants of Corruption: Interaction effects ................................................................ 49
Chapter Five: Conclusions & Reflections ............................................................................................... 53
5.1 Conclusions .................................................................................................................................. 53
5.2 Critical Reflections ...................................................................................................................... 53
References ............................................................................................................................................. 55
Chapter One: Introduction
1.1. Background of the Study
In recent years there is a growing consensus on the notion that bad governance and low
quality institutions underlie the poor performance of many economies and condemn many
people to languish under the yoke of extreme poverty. A problem closely related to these
concepts is corruption. Even though corruption occurs in all countries, it is generally accepted
that it is often more widespread and its effects are more detrimental in developing countries
(Olken and Pande, 2012). According to the World Bank, estimated bribe payments worldwide
amount over one trillion dollars and 25% of the GDP of African countries is lost due to
corruption each year (cited in Sequeira, 2012). Transparency International, the largest
worldwide anticorruption non-governmental organization, emphasizes the costs of corruption
as follows:
.............The cost of corruption can be divided into four main categories: political, economic, social and
environmental. On the political front, corruption is a major obstacle to democracy and the rule of law. In a
democratic system, offices and institutions lose their legitimacy when they’re misused for private advantage.
.............. Economically, corruption depletes national wealth. Corrupt politicians invest scarce public resources in
projects that will line their pockets rather than benefit communities, and prioritise high-profile projects such as
dams, power plants, pipelines and refineries over less spectacular but more urgent infrastructure projects such as
schools, hospitals and roads. Corruption also hinders the development of fair market structures and distorts
competition, which in turn deters investment. Corruption corrodes the social fabric of society. It undermines
people's trust in the political system, in its institutions and its leadership. A distrustful or apathetic public can then
become yet another hurdle to challenging corruption. Environmental degradation is another consequence of
corrupt systems. ........... precious natural resources are carelessly exploited, and entire ecological systems are
ravaged. From mining, to logging, to carbon offsets, companies across the globe continue to pay bribes in return
for unrestricted destruction
(Transparency International, 2015).
Recognizing these adverse effects of corruption on society, international organizations such as
the World Bank, the United Nations and Transparency International etc. have called for
widespread reforms in developing countries. These organizations and individual countries
have promoted various anticorruption policies by committing large amounts of resources to
fight against corruption. In order to be able to design appropriate and effective anticorruption
policies, we need to clearly know at least two things related to corruption: the determinants of
corrupt behaviour and its consequences on development.
1
For many years, corruption was considered primarily a political problem to be studied in
political science with little or no implication to economic development. Besides, most
international financial institutions refrained from working on corruption as it was regarded as
violating the principle of non-interference in the political affairs of their members (Soreinde,
2014). Since early 1990s, however, many studies have clearly shown the relation between
corruption and governance issues, on the one hand, and economic development, on the other.
Moreover, it has become clear that, if done right, governance issues may be addressed without
interfering in the political affairs of countries (Ibid.). Partly as a result of these changes, there
has been a tremendous increase in the number of theoretical and empirical works on the
economics of corruption. A major problem related to studies on corruption is the difficulty of
measuring the actual magnitude of corruption. This has in fact for so long limited the ability
of economists and political scientists to provide a menu of effective anticorruption policy
intervention tools. Because corruption is illegal or at least immoral, corrupt transactions take
place secretly and this makes measuring their occurrence very difficult (Serra and
Wantchekon, 2012). Until recent years, most of the studies on the determinants and effects of
corruption were basically macro studies, focused at national levels. Because of the difficulty
of objectively measuring corruption, these studies rely on aggregate data estimating
corruption using survey based indices to explain cross-national differences of corruption
perceptions (Olken and Pande, 2012). Based on the findings of these studies, most of the
anticorruption policies focused on changing or reforming the political institutions and the
legal systems and targeted national or regional bureaucrats often with unsatisfactory results in
curbing corruption. Measuring the corrupt behaviour of individuals using surveys has
disadvantages. One of these disadvantages is the high probability of measurement errors in
such measures. Besides, respondents are more likely to hide the truth when asked about their
involvement in corrupt practices for fear of legal and social sanctions (Serra and Wantchekon,
2012). Besides, there is evidence that corruption perception may significantly differ from
actual corruption (see Sequeira, 2012).
In recent years, microeconomic studies on corruption are growing. Some of these studies have
used lab-experiments to identify the micro-determinants of corruption (for a discussion, see
Armantier & Boly, 2012). And more recently, the remarkable progress made by economists in
using ingenious ways to measure corruption has led to the rise of well-identified,
microeconomic studies on corruption (Olken and Pande, 2012). A strand of these studies
2
focuses on identifying the determinants of corruption and thereby exploring avenues to curb
the incidence of corruption via various policy measures. Another strand focuses on the effects
of corruption on economic performance (Beekman et al., 2013). Most of the cross-national
studies on the determinants of corruption have emphasized the role of the political, economic
and social environments in explaining the variation in the levels of (perceptions of) corrupt
practices among nations or regions (see Lambsdorff, 2005; Treisman, 2000 & 2007; Soreinde,
2014; Seldadyo and Haan, 2006; Serra, 2006). Accordingly, these studies regarded structural
factors such as the overall level of development, religious traditions, political regimes or
design of legal systems as important determinants of the level (perceptions) of corruption.
But, recent studies show that short-term factors, that can possibly be manipulated by policy
interventions, may also influence corrupt behaviour (Beekman et al., 2014).
Early studies on the effects of corruption provided ambiguous results and debated on the
effects of corruption on development. Some of these studies indicate that corruption may
foster efficiency and hence growth while others argue to the contrary. But, in recent years
many studies generally indicate that corruption is a hindrance to growth and development
(Beekman et al., 2013). The World Bank states that even though countries with systemic
corruption may at the same time show strong economic performance, experience suggests that
corruption is bad for development (The World Bank, 2013). Corruption can hinder growth
and development through various mechanisms. Many macro studies and more recently,
micro-studies have investigated these mechanisms. A sizable number of macro studies show
that corruption acts as a distortive tax (reducing the marginal product of investment) and
hinders investment and economic growth, decreases foreign direct investment (FDI) and
exacerbates income inequality (see Bates, 1981; Campos et al., 1999; Lambsdorff, 2005;
Mauro, 1995; Schleifer and Vishny, 1993). Other studies also show that it lowers levels of
human capital and thus decreases the flow of FDI and growth by hindering the supply of
public services such as education and health care (Lambsdorff, 2005). Studies also show that
it may lead to the emergence of inefficient firms or use of low technologies by firms to
enhance bargaining (Svensson, 2003). Recently many micro-level studies also confirm most
of these findings. These micro studies indicate that corruption reduces incentives to invest, to
engage in profitable economic activities, to contribute to public goods and severely affects the
ability of governments to provide infrastructures, education and redistribute wealth among
their citizens in developing countries (Beekman et al., 2013; Beekman et al., 2014; Olken,
2006, 2007; Reinikka & Svensson, 2004 ).
3
In this study, we investigate the micro-determinants of corruption in Liberia. Despite some
progress in recent years, corruption is a big problem in Liberia. The country ranked 94 out of
175 countries in the Corruption Perception Index (CPI) in 2014 (Transparency International,
2015). As one of the poorest countries in the world (ranked 175/187 in the HDI in 2014),
corruption is regarded as one of the challenges to development in Liberia. Previous microlevel studies indicate that chiefs in rural Liberia engage in ‘corrupt’ practices (see Beekman et
al., 2013; Beekman et al., 2014; Beekman and Bulte, 2015). For example, Beekman et al.
(2014) identify two mechanisms by which villages chiefs may engage in corrupt activities:
using communal resources for own benefit and appropriating private property of specific
community members. This study aims to identify the determinants of corruption of village
chiefs in rural Liberia using a micro dataset. We investigate corruption behaviour at local
level among village chiefs based on a novel data collected from 132 communities using an
objective way of measuring corruption.
1.2. Statement of the Problem
As stated earlier, in order to effectively combat corruption with appropriate policy tools, at
least two things should be clearly understood. First, there is a need to understand the
underlying incentives and drivers of corrupt behaviour that could explain why some leaders
choose to engage in it while others choose not to. Second, the ways and if indeed corruption
hinders growth and development should be empirically investigated. There is little doubt that
in both of the above issues current economic research is lagging far behind where it needs to
be to offer policy makers with the appropriate policy toolkit to effectively combat corruption.
While many macro studies based on cross-national comparisons of indices of corruption
perceptions have attempted to tackle these two issues, there are only few well identified
micro-level studies on corruption that rely on objective measures. This is mainly due to the
difficulty of collecting accurate and objective data on (direct) individual corruption decision
making. As a result, there are so few hard data sets relating to micro-level corruption which
tend to be highly context specific and relate to the “victims” rather than those who commit
corruption (Abbink and Serra, 2012). This is a major setback in the efforts to fight corruption.
If we are not able to accurately and reliably measure corruption, we cannot properly identify
its extent and magnitude. This in turn implies that we cannot meaningfully test theories of
‘causes’ corruption against the data to help us understand the fundamentals of how and why
4
corruption occurs, and effective anticorruption strategies cannot be formulated, tested and
adapted to different settings (Sequeira, 2012).
Until very recently, most studies on the determinants and consequences of corruption were
essentially macro studies based on cross-national comparisons of indices of (perceptions of)
corruption. These studies have two basic limitations. First, they are cross-sectional studies and
hence do not enable us to tell a convincing causal story. It is difficult to establish causal
relationships using observational data that is prone to endogeneity bias (Seldadyo and Haan,
2006). The correlations can go both ways and it is not clear if some of the variables or factors
are causes or consequences of corruption. For example, countries with lower (perceptions) of
corruption may be expected to grow faster but on the other hand the causality can go in the
reverse direction: countries with high income may be expected to have the necessary
resources and good quality institutions that will enable them to combat corruption effectively.
Moreover, as cross-sectional studies, their results vary depending on the model specifications.
Variables that were significant in some studies become insignificant in others when the model
specification changes or other relevant variables are included. Besides, by taking countries as
study units, these studies mostly emphasize structural causes and ignore the role of the
individual decision maker who should be the subject of study when analysing concepts like
corruption. Second, they are based on measuring corruption using subjective indices of
perceptions of corruption not on the actual occurrence of corruption and actual levels of
corruption could vary significantly from perceptions of corruption for many reasons (see
Sequeira, 2012). These perception indices are provided by private firms or international
organizations based on surveys of country citizens, business analysts, risk analysts and
experts and as such these values are far from being objective (Serra, 2006). Besides, they may
be measured incorrectly. Effective anticorruption policies cannot be designed based on the
empirical findings of these studies ( see Abbink and Serra , 2012, pp. 78-79 for a detailed
explanation).
In recent years, lab-experiments are being used to investigate the micro-determinants of
corruption and thereby probe possible effective ways of curbing corruption. Despite their
advantages in enabling the researcher to directly observe individual decision making in
engaging in corruption, lab-experiments have low external validity; i.e., they are less likely to
represent individual decision making in a real environment (See Armantier and Boly, 2012 for
a discussion). In this regard, field experiments offer an alternative to lab-experiments and they
are relatively closer to what happens in a real world environment than lab-experiments.
5
Micro-level investigations of the determinants of corruption based on objective measures are
highly needed if we are to make progress in understanding this seemingly familiar but
completely complex and multi-faceted concept.
Particularly in Africa, even though bad governance is generally regarded as a hindrance for
development, only few studies focus on the role of governance at the local level (Turley et al.,
2013). A majority of the population in Africa lives in rural areas, where the national state has
limited reach, and hence the quality of local governance may greatly affect development
outcomes (Acemoglu et al.,2013).The lowest tier of government in most sub-Saharan African
countries is occupied by traditional rulers or chiefs who play many significant roles in the
socio-economic activities of their communities. These include, among others, raising taxes,
allocating resources (including land) and controlling the judicial system in rural areas. Despite
their diverse roles in the socio-economic aspects of African societies, relatively little is
known about how chiefs exercise their power and the effects of constraints on their power on
economic development (Ibid.). This is rather disappointing because the majority of the
population in Africa live in abject poverty in rural areas despite many national and
international efforts to alleviate poverty and to reduce income inequality with urban areas. In
this regard, it is highly important that we understand the local determinants of corruption
(often a symptom of bad governance) of village chiefs in rural areas in Africa.
Recent micro-level studies in rural Liberia indicate the existence of significant corruption
among villages and its negative effects on development (Beekman et al., 2013; Beekman et
al., 2014; Beekman and Bulte, 2015). In particular, these studies find that members have
reduced incentives to engage in productive activities, to invest and to contribute to public
goods in villages with corrupt chiefs. Accordingly, these studies also suggest that these
negative effects of corrupt chiefs can undermine the performance and success of development
projects as such projects often require complementary private investments and public
mobilizations (Beekman et al., 2014). To the best of our knowledge, there are no studies that
investigated the determinants of the corruption behaviour of the village chiefs in rural Liberia.
In this study, we try to provide a simple theoretical framework of modelling the determinants
of corruption of village chiefs in rural Liberia using the standard incentives theory of
Economics. We will complement our theoretical framework with a sociological perspective,
particularly with the social learning theory developed to explain deviant behaviour. This is
6
important because we would expect that the decision to engage in corrupt practices may also
depend on intrinsic motivations and the value framework within a society in addition to the
incentives framework and structural factors. We use a more objective way of measuring
corruption among village chiefs based on a field experiment done in rural Liberia.
Understanding the determinants of corrupt behaviour using basic (economic) concepts may
offer significant guidance on how to design effective anticorruption policies( Olken and
Pande, 2012). This study seeks to contribute to the growing but still relatively scanty microlevel evidence on the determinants of corruption in rural Africa.
1.3. Objective of the Study and Research Questions
The general objective of this study is to identify the micro-determinants of corruption of
village chiefs in rural Liberia. More specifically, the study attempts to address the following
two objectives. First, it examines the relation between chiefs’ characteristics and their corrupt
behaviour. Second, it investigates if the corrupt behaviour of chiefs responds to economic and
socio-cultural factors as well as to the presence of monitoring and the threat of punishment.
In line with these objectives, the research is guided by the following research questions:
1. Do characteristics of the chief determine his/her corrupt behaviour?
2. Which socio-economic and cultural factors affect the corruption of the chief?
The study is significant in that it enables us to understand how the opportunities and incentives
for corruption take place in development projects and reflect on the lessons this knowledge may
offer for policy formulations for combating corruption.
1.4. Organization of the Thesis
The thesis contains five chapters. The remainder of the thesis is organized as follows. The
second chapter introduces the theoretical framework and reviews the empirical literature
related to the determinants and impacts of corruption. It also provides a brief account of the
definition and types of corruption as well as the different types of recently emerging measures
of corruption. A brief account of the history of leadership and corruption in Liberia is also
7
included in this chapter. Chapter 3 presents the data sources, model specification and
estimation strategy, description of variables and forwards the hypotheses. Chapter four
addresses the results and discussions. The last chapter concludes the study and provides the
implications of the results and critical reflections on the study.
8
Chapter Two: Literature Review and Theoretical Framework
2.1.
Definition and Measures of Corruption
2.1.1. Definition of Corruption
Even though corruption is not a new phenomenon, it is hard to give a precise and universally
accepted definition (Aidt, 2003). Corruption is influenced by a multitude of factors including
economic, social, political, legal, cultural and historical and involves various officials at
different hierarchies of leadership. For the purpose of this study, we employ the definition of
corruption suggested by Rose-Ackerman (1993, p.1) and is widely used by many international
organizations such as the Transparency International as well as by many researchers. The
definition states that corruption is “the misuse of public office for private gain.” While what
constitutes ‘misuse’ is ambiguous, the definition recognizes that sometimes public office can
legitimately provide private benefits to politicians and bureaucrats (Rose-Ackerman, 1993).
What is considered ‘misuse’ will most likely vary in different societies and contexts. But still,
we believe that this definition is relevant to our study of corruption in rural Africa due to the
wider scope and potential for ‘misuse’ of power by village chiefs given their many roles in
the socio-economic aspects of their communities. Corruption is multidimensional and can
take a variety of forms (see Soreinde, 2014 for a glossary of corruption). We will provide the
definitions of only three types of corruption for the purpose of our study. “Grand corruption
consists of acts committed at a high level of government that distort policies or the central
functioning of the state, enabling leaders to benefit at the expense of the public good while
petty corruption refers to everyday abuse of entrusted power by low- and mid-level public
officials in their interactions with ordinary citizens, who often are trying to access basic goods
or services in places like hospitals, schools, police departments and other agencies”
(Transparency International, 2015). When corruption is so prevalent that it is part of the
everyday structure of society, we have what is called systemic corruption. According to Serra
and Wantchekon (2012, P.1) all acts of corruption have two common features: (1) “they all
rely on rule breaking on the part of public officials for the achievement of some form of illicit
private gain and (2) they all take place behind closed doors.” In order to effectively administer
public offices, there is a need to delegate power and authority and civil servants must be
trusted to make decisions on multiple issues. This discretionary authority may lead to
temptations and opportunities to pursue personal interests at the expense of what may be best
for society (Soreinde, 2014). When this happens, it is corruption.
9
2.1.2 Measures of Corruption
As stated earlier, the primary challenge faced by scholars of corruption is the difficulty of
measuring (actual) corruption. According to Sequeira (2012), measuring corruption is difficult
for a number of reasons. First of all, corrupt officials seek to hide their behaviour for fear of
punishment and shame. Corruption can also take many forms and involve many different types
of public officials. And as such “identifying the full extent of different types of corruption will
often require a deep understanding of the broader institutional, social, political and cultural
context in which corrupt behaviour is taking place. The main challenge is to identify objective,
replicable and yet adaptive measures of corruption, capable of capturing new developments in
corrupt behaviour as they emerge” (Sequeira, 2012, p. 147).
Until very recently, most measures of corruption were based on surveys of perception. While
these perception surveys have the advantage of large coverage, they are subjective and may not
measure corruption accurately (Olken and Pande, 2012). Perception based measures have
several methodological problems, such as unpredictable sampling and reporting bias (see
Sequeira, 2012; Triesman, 2007). “At a more practical level, the methodology and degree of
completeness of these indices varies both from country to country and from year to year,
rendering cross-national and longitudinal studies hard to interpret”(Sequeira, 2012, p. 150). The
most commonly used perception indices are the Corruption Perception Index (CPI), a survey of
experts and public opinion on corrupt practices in over 150 countries and the Bribe Payers
Index (BPI), a survey of businessmen and investors on the likelihood of having to pay a bribe
in over 60 countries. Both of these indices are
produced annually by the Transparency
International. The International Country Risk Guide (ICRG) and the World Bank Governance
Indicators (WBGI) also yearly publish
indices of the quality of institutions in different
countries including perceptions of petty and grant corruption as well as the degree of state
capture by elites and private interests (Ibid.).
In recent years, economists have shown a remarkable improvement in their ability to measure
corruption (Olken and Pande, 2012). Scholars have employed various direct and indirect
techniques of measuring corruption. Among the direct techniques are for example; documented
measures of corruption (bribes), observing corruption in the field and transaction-level data of
bribe payments. Among the indirect techniques are: forensic economic approach or estimates
10
from market or statistical inference and minding gaps in the data (see Sequeira, 2012; Olken
and Pande, 2012 for an elaborate discussion of these methods). Among these, the method that is
increasingly becoming popular to measure corruption in the field is what is commonly called
‘minding gaps in the data’. This method uses the difference between two data or measures that
suggest ‘illicit’ behaviour as an indicator of corrupt behaviour(Beekman et al., 2013). The
difference might be identified in case of discrepancies between different official data sources;
between administrative data and results from an independent household survey (see for
example, Olken, 2006, 2007; Reinikka and Svensson, 2004) or between two primary sources of
data (see Beekman et al., 2013). Olken and Pande (2012) describe this approach also as ‘graft
estimation by subtraction’ and is regarded as the most commonly used approach for estimating
graft (i.e., the theft of government funds).
For the purpose of this study we employ a form of ‘minding the gap’ measure of corruption
following Beekman et al. (2013, 2014) in their first studies of corruption focusing on African
smallholder farmers. The researchers partnered with a development project that delivers
agricultural inputs to some villages in rural Liberia. These inputs include -rice seeds, vegetable
seeds and agricultural tools1. Before the inputs were taken to the villages, they were carefully
measured (‘ measure 1’). They then asked the chief to keep these inputs in his house for two
days. After two days the inputs were measured again to check if some inputs were missing or
‘stolen’ by the chief (‘measure 2’) before they are distributed to beneficiaries in the
community. The difference between these two measures is interpreted as an indicator of the
chief’s tendency to ‘misuse’ his/her power to appropriate community resources for ‘private
gain or use’ (‘corrupt practice’ following the definition employed in our study). They then used
this difference between the two measures to construct various proxies for corruption. Such a
measure has certain advantages: it does not suffer from a bias resulting from incompetent
bookkeeping and is not based on survey responses which are more likely to be biased
(Beekman et al., 2013). In their study, they find that seeds were diverted in 36% of the
communities and diverted inputs were observed in almost half of the communities. This
suggests a high prevalence of ‘misuse’ of power or ‘corruption’ of village chiefs in rural
Liberia. But, Beekman et al. are adamant that alternative explanations may exist for the ‘stolen’
inputs and their measure may not be a perfect measure of actual corruption (See Beekman et
al., 2013, 2014; Beekman and Bulte, 2015). Even if this measure may be noisy and may not
1
Seeds: 25 kg upland or paddy rice (depending on land type); 3 kg beans and peanuts; 5 kg corn; 20 g pepper
seed; 5 g bitterball seed. Tools: cutlasses and regular hoes (4 each); shovels, files and watering cans (2 each).
11
arguably reproduce exactly a real corruption environment, we believe that it is a more objective
and reasonable proxy for corruption given the difficulties of observing actual corruption. In our
case, the distributed inputs are only rice seeds which were delivered in 2012 as part of a large
scale development project aimed at poverty alleviation in rural Liberia. A 30 Kg bag of rice
seeds was kept at each chief’s house for 2 days and then was measured again to check if any
amount was missing. Then, we used the amount of rice missing as a proxy for the corruption
behaviour of the chief. The implicit assumption here is that chiefs who stole the rice seeds in
such a situation are also most likely to misuse their power for illegitimate personal benefits
through diverting community resources or any other means (Beekman & Bulte, 2015).
2.2.
Theoretical Framework and Empirical Literature
2.2.1. The Incentives for Corruption and the Individual Decision
Maker: An Economic Perspective
Why do some officials or leaders tend to misuse their public power for personal gains while
others do not? Why is corruption more widespread in some regions or countries than others? A
commonly used conceptual framework by political economists to analyse the drivers of
corruption is the incentives theory. We employ the insights from the theory to guide our
analysis in the search for the local or micro-level explanation for the corrupt behaviour of
village chiefs in rural Liberia. The incentives theory postulates that a behaviour or an action is
pursued if its expected benefits are greater than the expected costs for an individual. In line
with this, factors that are expected to increase the (perceived) expected benefits and/or reduce
the expected costs of an action(behaviour) tend to drive the action(behaviour). In the same
manner, factors that are expected to decrease the (perceived) expected benefits and/or increase
the expected costs of an action(behaviour) tend to discourage it. Most (economic)studies on the
drivers of corruption (at an individual level) posit that individuals engage in corrupt activities
when the expected benefits of doing so are high and/or the expected costs of doing so are low.
So, the decision to engage in corrupt practices can be analysed using incentives theory that
explains corruption as the result of a rational choice. According to the theory of individual
utility maximization, an individual will be involved in corruption if the expected benefits of
doing so are greater than the expected costs (Soreinde, 2014). The (expected) benefits of a
corrupt act possibly are monetary gains as well as positions and power for oneself, family or
relatives. The expected costs of a corrupt act can be broadly divided in to psychological, social
12
and financial costs. More specifically, these costs include the bribe payment, moral “costs” of
breaking norms and rules, efforts to hide the crime and money laundering, as well as the
perceived risk of detection and the consequences of prosecution and punishment. Indirect costs
could also be loss of reputation if the corruption is revealed, including a loss of status and
future income or benefits, for example as result of losing work positions or tenders (Ibid.).
Using this conceptual framework, we can categorize the determinants of corruption into two:
those factors that are assumed to increase the (perceived) expected benefits of engaging in
corrupt activities and those that are assumed to decrease the expected costs of engaging in those
activities. The factors that are expected to affect the expected benefits and/or costs of
corruption depend on many economic, socio-cultural, institutional and political conditions. For
example, using the logic of the incentives theory, Olken and Pande (2012) developed a simple
model to look into the incentives bureaucrats face when making decisions to engage in corrupt
activities. They assumed the bureaucrat receives a wage w from the government and can
receive an outside option v elsewhere, if fired from the job. The bureaucrat can decide to be
corrupt or honest. If he chooses to be corrupt, he is detected with a probability p, is fired, and
receives the outside option v. If his corrupt act is undetected, he receives his wage w plus the
bribe b, but he also incurs a dishonesty cost d. In such a situation, Olken and Pande (2012)
postulated that in equilibrium, the bureaucrat will be corrupt if and only if 𝑤 − 𝑣<
1−𝑝
𝑝
(𝑏 −
𝑑). This implies that the lower the wage (w) of the bureaucrat, the higher the perceived benefit
(utility) of engaging in corruption (due to the potential for higher need for additional income)
and hence the more likely the bureaucrat will become corrupt. In the same manner, the
presence of rents (a higher b) tends to increase the expected benefits of engaging in corruption
and hence will reinforce corrupt behaviour. This framework suggests the following options for
reducing corruption: increasing the returns to staying on the job (w), or, equivalently decreasing
the outside option (v) by increasing punishments or increasing the probability of detection (P)
(Ibid.).
Olken and Pande (2012) state that there are costs of engaging in corruption even when it is
undetected which they referred to as the dishonesty cost(d). The dishonesty cost refers to the
psychological and social costs of engaging in corruption. This cost depends on individual
beliefs and preferences as well as socio-cultural factors which determine the intensity of the
sense of guilt & social sanctions faced by the corrupt individual (Ibid.).These may include, for
13
example, how corruption is defined by the society, how wide spread it is in the society
(whether there is systemic corruption) and other cultural factors. In some societies giving bribe
is considered just a gift to promote future business deals or an additional compensation in
return for receiving some favours or for services rendered (Varese, 2000 as cited in Abbink and
Serra, 2012). Some societies regard the graft or embezzlement of public resources by their
leaders as mere compensations for the services they provide to the community (also works for
nepotism, clientelism where in some societies are accepted norms). So, in such societies the
sense of feeling of guilt as well as the social sanctions on corrupt individuals will be minimal or
non-existent. If there is systemic corruption, then society will condone corruption and no
sanction or action will be taken against those who are engaged in it. All these factors reduce the
dishonesty cost of engaging in corrupt activities in any given society and hence drive corrupt
behaviour. According to Olken and Pande (2012), if potential bureaucrats (leaders) differ in
their dishonesty cost , those who have lower dishonesty cost will self-select to become leaders.
“People with a lower dishonesty cost will have a higher utility from becoming bureaucrats than
those with high dishonesty cost, as they will be relatively more efficient at corruption.” (Olken
and Pande, 2012, p. 496 ). So, according to them, this may make fighting corruption harder
because a corrupt system may attract bureaucrats who are more prone to corruption.
As stated above, the decision of the bureaucrat to engage in corrupt activities is also influenced
by factors that affect the expected costs of engaging in corruption. One of these factors is the
probability of being detected p (Olken and Pande, 2012). The threat of the probability of being
detected depends on the intensity of monitoring and punishment which in turn is influenced by
the availability, access to and flow of information (transparency), the quality of institutions and
the political will and commitment to fight corruption. This implies that in societies
characterized by lower flow of information (lack of communication), lower quality of
institutions and lack of political will and commitment; the perceived threat of detection and
punishment will be lower. Hence, this will encourage the decision to engage in corrupt
activities. Another factor that influences the expected costs of a corrupt act is the probability of
being fired or losing one’s position upon detection. The bureaucrat may have the chance to
receive an outside option v if he is detected in corruption and is fired. The probability of being
fired when detected in fact depends partly on the nature of the political decision making system
and the presence of competition. If the leader has no potential competitors or opponents and is
entitled to the position, then the threat of losing the power or position will not exist or at least
will not be credible. So, in such a situation there will not be an incentive to behave non14
corruptly to keep ones job or position. This partly explains why authoritarian governments have
little responsiveness to public pressure against corruption (at least in the long term). On the
other hand, if there is competition for a position, this will create a disciplining incentive for the
incumbent to act in a non-corrupt manner so as to be elected in the future. Countries with more
political competition have stronger public pressure against corruption- through laws,
democratic election and even the independent press and so are likely to use government
organizations that reduce rather than maximize the benefits of engaging in corruption (Ibid.).
In a related argument, Schleifer & Vishny (1993) claim that the structure of government
institutions and the nature of the political process can be vital determinants of the level of
corruption. In particular, weak governments that have less control over their agencies
experience very high corruption levels. This often holds true for fragile states which are in postwar transition like Liberia. Countries affected by war and conflict are often recognized as
highly prone to widespread corruption: they usually have weak administrative and judicial
institutions and low capacity to monitor and enforce rules against corruption. Social divisions
as a result of the war may weaken shared conceptions of the public good and social norms that
could otherwise hinder corrupt behaviour (Cheng and Zaum, 2011). Many studies indicate that
post-war communities are typically characterized by reduced community cohesion and hence
lower trust (For example see Richards et al., 2005 ) and this may lead to more corruption. The
outside option a corrupt leader receives (v) when fired also depends on the presence and/or
level of punishment (Olken and Pande, 2012). This in turn is determined by the level of
monitoring, the power of the monitoring authority to persecute corrupt individuals and the
independency of the courts to enforce the punishments. The effectiveness of increased
monitoring in reducing corruption in practice depends on the nature of the very individuals
tasked with monitoring and enforcing punishments as well as the possibility of getting enough
evidence to make the case (which is also related to access to information or transparency and
political will). If these individuals are themselves corruptible, increasing monitoring may not
help much and may rather simply increase transfers from low-level officials to auditors (Ibid.).
In short, our conceptual framework based on the incentives theory postulates that the
propensity to be involved in corruption depends on rational (in the neoclassical sense of the
term) assessment of (expected) benefits and (expected ) costs. One may question the power of
our theoretical model to predict corrupt behaviour since we assumed rational and informed
decision makers. It may be argued that the power of our model to explain the determinants of
15
corrupt behaviour will be limited by variation in individuals’ perceptions of costs and benefits
associated with engaging in corruption, as well as varying levels of rationality. And, indeed
individuals in reality often make irrational decisions due to the many intellectual limitations we
have. But according to Soreinde (2014), even if the assumption of rational and informed
players seems not reasonable, the concept is more useful than theories that assume more
random decision making or decisions steered by framework conditions. Moreover, she states
that it is in line with the fundamental role of the rationality assumption in criminal justice
systems. Placing blame and sentencing would only make sense if individuals are assumed to
steer their own decisions. Besides, recent behavioural research shows that that many irrational
choices or behaviours of individuals follow a pattern (for example high rate of time preference,
high risk or loss aversion). This implies that theories explaining the incentives to take part in
corrupt acts can be adjusted based on irrational decision-making processes and their validity
can be quite high despite natural intellectual shortcomings (Ibid.).
2.2.2. A Sociological Perspective on the Drivers of Corruption: The
Social Learning Theory
Sociologists and Psychologists often use the so called social learning theory developed by
Akers (1979) to explain deviant behaviour(also called ‘problem behaviour’) of individuals or
groups. The social learning theory is based on four interrelated, overlapping and mutually
reinforcing concepts that operate to promote or undermine conformity: differential Association,
differential reinforcement, definitions and imitation (Tavits, 2005). Differential association
refers to interaction and identity with different groups (Akers et al., 1979). It implies that
through interaction with others, individuals learn the values, attitudes, techniques and motives
for criminal behavior. Differential reinforcement is the implementation of reinforcing only the
appropriate response (or the behavior you wish to increase) and applying extinction to all other
responses. Extinction is the discontinuing of a reinforcement of a previously reinforced
behavior. Problem behaviour is behaviour that is socially defined as a problem, as a source of
concern, or as undesirable by the social and/or legal norms of conventional society and its
institutions of authority; it is behaviour that usually elicits some form of social control
response, such as a statement of disapproval, or even incarceration(Donovan et al., 1991).
Definitions generally refer to norms, attitudes and orientations toward certain behaviour while
imitation refers to the modelling of others’ behaviour (Tavits, 2005). Akers et al. (1979, pp.
637-638) state that:
16
.......Social behaviour is acquired both through direct conditioning and through imitation or modelling of
others' behaviour. Behaviour is strengthened through reward (positive reinforcement) and avoidance of
punishment (negative reinforcement) or weakened by aversive stimuli (positive punishment) and loss of
reward (negative punishment). Whether deviant or conforming behaviour is acquired and persists depends on
past and present rewards or punishments for the behaviour and the rewards and punishments attached to
alternative behaviour-differential reinforcement. In addition, people learn in interaction with significant
groups in their lives, evaluative definitions (norms, attitudes, orientations) of the behaviour as good or bad.
..............The more individuals define the behaviour as good (positive definition) or at least justified
(neutralizing definition) rather than as undesirable (negative definition), the more likely they are to engage in
it.
Accordingly, social learning theory can offer a possible alternative insight into individual
motivation to engage in corrupt behaviour (Tavits, 2005). The basic mechanism of the social
learning theory, according to Akers (1998), works as follows: behaviour is acquired and
sustained through three mechanisms, adopting definitions favourable to illegal behaviour via
differential association with one’s peers, imitating such behaviour by peers, and the positive
reinforcement provided by rewards for such a behaviour (as cited in Tavits, 2005). As stated
above, the more a behaviour is defined as good (“positive” definition) or as justified
(“neutralizing” definition) rather than bad (“negative” definition), the more likely individuals
are to engage in it (Akers, 1979; Akers, 1998 as cited in Tavits, 2005). Many empirical studies
also indicate the existence of strong and consistent relationship between individual’s definitions
and attitudes toward the behaviour in question and misbehaviour in a variety of contexts such
as substance abuse, sexual aggression, white collar crimes, and police misconduct including
police corruption (Krohn et al., 1985; Akers 1998 as cited in Tavits, 2005).
Drawing from the social learning theory, Tavits (2005) posits that individual level motivation
helps to understand why people are willing to engage in corruption (which can be regarded as a
problem or deviant behaviour in many societies). She took the individual level approach in her
study and tackled the question of why individuals engage in corrupt practices. She argued and
showed that individuals will be more likely to engage in corrupt behaviour when they do not
define corruption as morally or situationally wrong, but rather define it positively or neutrally
(or consider it an acceptable mode for exchange). Furthermore, people are more likely to
engage in corrupt activities if they perceive that corruption is widespread and, thus, approved
(Tavits, 2005). Based on these arguments, Tavits (2005) argue that structural reasons
emphasized by most economic studies of corruption can only be secondary causes of
corruption. For example, if people believe that corrupt behaviour is normal or beneficial,
institutional or structural changes will not help that much to curb corruption. In a related
17
argument, Uslaner (2004) suggested that the corrupt behaviour of individuals may be resistant
to large-scale structural reforms making micro-level analysis even more important (as cited in
Tavits, 2005). Furthermore, in their study on corruption in Eastern Europe, Miller et al. (2001)
argue that values, i.e. condemnation of bribery, increased resistance against corruption(as cited
in Tavits, 2005).
Individuals also develop and sustain behaviour through conditioning and imitating or modelling
of others’ behaviour (Tavits, 2005). If a problem behaviour is perceived to be widespread and
approved, then individuals are more likely to engage in such behaviour. Definitions and
imitation mutually reinforce each other in their effect on individual’s behaviour. By taking
others’ behaviours as models, a person learns how to define a behaviour as ‘good’ or ‘bad’,
‘right’ or ‘wrong’. In short, individuals learn from their own past behaviour and from their
association with peers, both of which influence their evaluative definitions and reinforce
imitation. Applying these arguments to situations of corruption, Tavits (2005) stated that the
theory of social learning would propose that willingness to engage in corrupt behaviour can be
expected to increase when one does not define corruption as morally or situationally wrong, but
rather considers it a justified and acceptable mode for exchange. Likewise, the theory proposes
that individuals willingness to engage in corruption is expected to increase when they have
been exposed to corrupt behaviour before or at least perceive that corruption is widespread and,
thus, approved. We believe that such a sociological perspective would complement our
theoretical framework and provide additional insights into the determinants of corruption since
corrupt behaviour is also likely to be influenced by non-economic intrinsic motivations (sense
of guilt and shame) in addition to extrinsic economic incentives. In fact, as De Graaf (2007)
rightly remarked integrative or interconceptual explanations are the most useful to explain
corrupt behaviour.
2.3.
The Opportunities for Corruption: A Brief Review of the Empirical
Literature on the Determinants of Corruption
The decision to engage in corruption must not only be seen as the result of individuals’
incentives and motivations but also be understood to be influenced by context and framework
conditions( i.e., the external factors that the individual cannot influence) (Soreinde, 2014). A
variety of characteristics of country’s economic, political, cultural and social systems that
might affect the expected costs, benefits, or both of engaging in corrupt activities have been
suggested by economists and political scientists. Some of these include characteristics of a
18
given country; its political situation, its society, history, norms and institutional qualities etc..
Soreinde (2014) explains some of these factors including: civil servants’ authority and how
bureaucracies are organized, exclusive access to information, law enforcement and sanctions,
absence of reaction against corruption, rationalizing corruption (i.e., society’s general opinion
on corruption; for example, low wages and lack of state legitimacy may lead to a general
tolerance of corruption), loyalty to non-state institutions like ethnicity or tribe and condoning
corruption by management.
Economic Factors
- Income per capita
- Development
- Trade openess
- State regulation
- Public sector wages,
dependence on natural
resources, foreid aid
Political/ legal
Factors
- Political rights, democracy,
freedom of information, mass
media diffusion, federalism,
electoral system, political
instability, rule of law or
independence of judiciary
Historical
Factors
Corruption
- Colonial history, legal
origin
Socio-cultural Factors
-Religion, ethnolinguistic fragmentation,
education of the population
Figure 1. (Macro-level) Determinants of Corruption
Many empirical macro-studies on corruption have searched for the economic and noneconomic determinants of corruption (Seldadyo and Haan, 2006). Most of these studies are
cross-sectional and rely on specifying regression models with a wide variety of explanatory
variables to find the ‘true’ determinants of corruption. As such, the findings of these studies
19
vary depending on the model specification. Often, a variable significant in a particular
specification of the model loses its significance when the model specification changes (some
other variables are incorporated). In other words, claims concerning the determinants of
corruption are conditional and the robustness of the findings is questionable. Recognizing this
limitation, some studies have used a sensitivity analysis2 (see Aidt, 2011) to find any truly
robust determinants of corruption among the various factors identified by empirical studies as
significantly related to corruption. For example, Serra (2006), using a global sensitivity
analysis based on the Leamer’s Extreme-Bounds analysis3, tested a total of 16 variables to find
robust determinants: four
economic
variables (economic development, openness to
international trade, state intervention in the economy, endowment of natural resources), five
socio-cultural variables (British legal system, British colonial heritage, Protestant religion,
ethnolinguistic fragmentation, education of the population) and seven political variables (base
political rights, uninterrupted democracy, freedom of information, mass media diffusion,
federalism, electoral system, political instability). She finds five of these variables to be
robustly related to corruption. These are economic development or income, political instability,
colonial heritage, protestant religion and uninterrupted democracy.
Treisman (2000), using a cross-national study, found six robust determinants of corruption.
Protestant traditions, histories of British rule, more developed economies and (probably) higher
imports and long exposure to democracy were robustly related to lower corruption while
federal states were perceived more ‘corrupt’. Treisman (2007) reviewed the cross-national
empirical literature (by political scientists and economists in the late 1990s and early 2000) on
the causes of corruption and examined the robustness of the reported findings. Although the
causal direction is usually ambiguous, he finds that, instrumenting with income as of 1700,
more developed countries are perceived less corrupt. However, when income is controlled, he
finds that most factors that predict perceived corruption do not correlate with recently available
measures of actual corruption experiences. He also finds that reported corruption experiences
correlate with lower development, and possibly with dependence on fuel exports, lower trade
openness, and more intrusive regulations. Lambsdorff (2005) reviewed the cross-national
studies on the consequences and causes of corruption and concluded (though the direction of
2
A sensitivity analysis involves a systematic evaluation of all possible linear (regression) models with
corruption as the dependent variable and a fixed number of potential determinants (typically 3 to 5) from the
target list of up to 75 as the explanatory variables, to establish which of the many potential determinants or
causes are robustly correlated with corruption.
3
(Leamer’s) Extreme-Bound analysis report an upper and lower bound for parameter estimates to examine the
sensitivity of parameters to model specification (see Seldadyo and Haan, 2006 for a detailed explanation).
20
causality is controversial) that corruption clearly correlates with a low GDP, inequality of
income, inflation, increased crime, policy distortions and lack of competition. With respect to
the causes of corruption, he noted that some of the empirical results were not consistent with
his expectations. For example, he finds the effect of higher official wages in reducing corrupt
behaviour to be rather limited. Also the impact of colonialism on corruption was ambiguous.
Press freedom and the (de facto) independence of the judiciary and prosecutors appeared to be
vital elements in reducing corruption. Complicated regulation of market entry and tariffs,
abundance of natural resources, distance to major trading centres were also found to lead to
increased corruption. He also finds that countries with high levels of democracy, or electoral
systems with high rates of participation, are able to reduce corruption but that the effect of
democracy is also not immediate but takes longer period. Thus, democracy reduces corruption
in the long run, but not the lukewarm (medium level) type of democracy. In terms of cultural
dimensions, hierarchical religions (the Catholic, Eastern Orthodox and the Muslim religions)
were found to increase corruption. As opposed to the findings of the study by Lambsdorff
(2005) and many other studies, Abbink et al. (2014) find that more market competition leads to
more bribes or corruption by crowding out altruism.
Seldadyo and Haan (2006) examined 70 economic and non-economic determinants of
corruption to find robust ones. Using factor analysis technique, they generated five new indexes
on the basis of these determinants: regulatory capacity, federalism, inequality, trade and
political liberty. Using two tests of Extreme Bounds Analysis, they find that one of these
indices, namely ‘regulatory capacity’ (the capacity of government to regulate and enforce law)
is the most robust determinant of corruption. But, using the Sala-i-Martin’s test4(see Aidt,
2011), they find that about 11-14 variables can be considered robust determinants. The other
robust determinants of corruption they find are: population density(-), Scandinavian legal
origin(-), ethnic tension(+), Socialism legal origin(+), portion of population with no religion(+),
ethnic conflict(+), illiteracy rate(-), government wage(+), sound money( area3 of Fraser
index)5(+), latitude(-), fuel export(+), primary school enrolment(+), external debt(-),
presidential(-), and portion of female in labour force(-). Table1 below summarizes the variables
or factors which were found to be robust determinants of corruption in the four studies
4
Sala-i-Martin’s test for robustness is that 95 percent of the cumulated density associated with the estimated
coefficients on the variable of interest (e.g., GDP per capita) across all the models considered should be on one
side of zero (also see Seldadyo and Haan, 2006 for a detailed explanation).
5
Sound money is a currency backed by a tangible commodity such as gold , silver or platinum. The more
common term is hard money.
21
described in this sub-section. For an extensive literature review of the empirical determinants of
corruption, see Seldadyo and Haan (2006). We believe that even though the structural causes
emphasized by the macro studies; the organizational structure and rules factors (like the role
model of leaders and managers) emphasized by meso-studies or the individual motivations
emphasized by micro studies provide a group of factors that explain corrupt behaviour, none of
these is alone sufficient to explain corruption. Hence, in this study we attempt to combine the
insights from these studies in our theoretical framework and test the resulting hypotheses
empirically. Accordingly, we will use some of the findings of these macro studies to suggest
and test possible micro-level explanatory variables to explain variation in corruption behaviour
in our study.
Table 1. Main Empirical Studies on Robust Determinants of Corruption
Author (s) and year
Robust variables or factors6
Approach and study design
Lambsdorff (2005)
Press freedom and the independence of
-Cross-sectional study
judiciary
-Reviewed previous studies
(-),
Complicated
market
regulations and tariff (+), abundance of
natural resources (+), distance to major
trading centres (+), long and high levels of
democracy (-), hierarchical religions (+)
Serra (2006)
Economic
development
or
income(-),
-Cross-national study of 62 countries
political instability(+), colonial heritage(+/-
-Used a global sensitivity analysis based on
),
the Leamer’s Extreme-Bounds analysis
protestant religion(-) and uninterrupted
democracy(-)
Treisman (2000)
Protestant traditions (-), histories of British
-Cross-national study
rule
development(-),
-Started by running a series of nested
(probably) higher imports (-), long exposure
(-),
economic
regressions beginning with only the most
to democracy (-), federalism(+)
exogenous
variables
and
progressively
including groups of variables, to check the
robustness of the results, he ran the same set
of regressions for four different corruption
indices
Treisman (2007)
Economic
development(-),
(Possibly)
dependence on fuel exports (+), trade
-Cross-national
-Reviewed previous studies
openness (-), more intrusive regulations (+)
Seldadyo and Haan (2006)
‘regulatory capacity’ (-)
Using two tests of Extreme Bounds Analysis
Population density(-), Scandinavian legal
Using the Sala-i-Martin’s test
origin(-), ethnic tension(+), Socialism legal
origin(+), portion of population with no
religion(+), ethnic conflict(+), illiteracy
rate(-),
money(+),
government
latitude(-),
wage(+),
fuel
sound
export(+),
primary school enrolment(+), external debt(-
6
+/ - indicate the sign of the effect of the variable on (perceived) corruption.
22
), presidential(-), and portion of female in
labour force(-)
In recent years lab-experiments are increasingly used to identify the micro-determinants of
corruption (Armantier and Boly, 2012). Such experiments are often done among university
students in a controlled environment and have certain advantages and limitations. One of the
advantages is that they allow the researcher to directly observe corrupt behaviour. A most
commonly cited limitation of lab-experiments is, arguably, their low external validity (See
Abbink and Serra, 2012; Armantier and Boly, 2012). Corruption studies based on labexperiments mostly focus on identifying the role of monetary and nonmonetary incentives in
influencing corruption behaviour. As such these studies investigate the effects of gender, age,
education level, (higher) wages, monitoring and punishment, whistleblowing and intrinsic
motivations (using framing effects) in promoting and/or deterring corruption behaviour ( See
Armantier and Boly, 20117; Armantier and Boly, 2012; Barr and Serra, 2009; Barr and Serra,
2010; Serra, 2011). For example, Armantier and Boly (2011) find that age, ability and
religiosity significantly affect corruption behaviour while they do not find a significant effect
of gender. They also find that (some level of ) monitoring and punishment can reduce
corruption, but may also crowd out intrinsic motivations for
honesty when intensified
(Armantier and Boly, 2011). Higher bribes were found to encourage corrupt behaviour but the
effect of higher wages in reducing corruption was ambiguous (Armantier and Boly, 2011).
Barr and Serra (2009), using a simple bribery game, find that when the game was framed as a
petty corruption situation and the negative external effects on innocent victims when bribes
change hands were high, private individuals were less likely to offer bribes. Higher negative
external effects were also found to be correlated with lower bribe acceptance, but they do not
find a significant effect of framing on the likelihood of accepting bribes (Barr and Serra,
2009). Barr and Serra (2010), in their study of corruption and culture, find that the values and
social norms that exist in the societies where individuals grew up partly influence their
tendency to engage in corrupt activities. Serra (2011), using a specifically designed bribery
lab-experiment, finds that combining top-down and bottom-up monitoring and accountability
system can effectively reduce corruption. So, some of these variables will be used and tested
in our study as potential micro-determinants of corruption.
7
This study combines lab and field experiments.
23
2.4.
A Brief Review of the Empirical Literature on the Consequences of
Corruption
Despite the general agreement in recent years on the adverse effects of corruption on
development, the exact relationship between corrupt leaders and the development performance
of their communities is not yet very clear. Some early studies on corruption indicated that
corruption may have beneficial effects on the growth of an economy. The most popular
argument in support of the beneficial effects of corruption rests on the so-called “grease the
wheels” hypothesis (Meon and Sekkat, 2005). This hypothesis was put forward by Leff (1964),
Huntington (1968) and Leys (1965) and states that corruption may have beneficial effects in a
second best world where distortions caused by ill-functioning institutions exist(as cited in
Meon and Sekkat, 2005).The argument is that an inefficient bureaucracy is an obstacle to
investment and that some corruption, acting as a “speed” or “grease” money, may help
overcome this obstacle by increasing efficiency. Acemoglu and Verdier (1998), for example; in
their study of property rights, corruption and the allocation of talent, considered an economy
where contracts are necessary to encourage investments. And they find that allowing some
level of corruption and enforcing property rights only partially may be optimal in such an
economy. Their reasoning is that; because preventing corruption is costly, partial property right
enforcement may be optimal, and preventing all corruption may be costly. As a result, they
suggest that it could be optimal for less developed countries, which may have less opportunities
for productive investment, to have an intermediate level of property right enforcement and
more corruption. More recently, Egger and Winner (2005), using a data set of 73 developed
and less developed countries, find that corruption is a stimulus for FDI, which is in line with
the position of Leff (1964), that corruption can be beneficial in overcoming regulatory and
administrative restrictions. In short, the “grease the wheels” hypothesis states that corruption
may help in overcoming bureaucracies, thereby raising efficiency hence investment and,
eventually, growth (Meon and Sekkat,2005).
But subsequent (macro) studies on the impacts of corruption have shown the negative impacts
of corruption on growth and investment (Bates, 1981; Campos et al., 1999; Mauro, 1995;
Schleifer and Vishny, 1993). These studies support what is called “sand the wheels” hypothesis
which states that although corruption may have benefits in countries with low quality of
governance, it may as well create additional negative effects(Meon and Sekkat, 2005). Aidt
(2009) took a critical look at the link between corruption and economic development and
24
concludes that the evidence supporting the ‘greasing the wheels hypothesis’ is very weak and
shows that there is no correlation between a measure of actual experience with corruption and
GDP growth. To the contrary, he finds a strong negative correlation between growth in genuine
wealth per capita (a direct measure of sustainable development) and corruption, supporting the
“sand the wheels” hypothesis. Many other studies also support this hypothesis (For e.g., Hines,
1995 & Wei, 2000 as cited in Aidt, 2009; Lambsdorff, 2005). Such a position is also reiterated
by many international organizations like the World Bank and Transparency International which
have stressed that corruption is bad for development.
At the micro level also, the empirical evidence on the impacts of corruption of leaders on
development is mixed (though at a lesser extent than the macro studies). On the one hand, a
sizeable body of literature finds that corruption hampers development; reduces incentives to
invest, to engage in profitable economic activities, to contribute to public goods and greatly
reduces the ability of governments to invest in human capital, infrastructures and to redistribute
wealth among their citizens in developing countries. For example, Beekman et al. (2014) find
that corrupt chiefs strongly undermine incentives to provide local public goods and to invest in
private goods by villagers in rural Liberia. In an earlier study in rural Liberia, they also find
that members, in communities with corrupt chiefs, have reduced engagement in economically
important income generating activities(Beekman et al., 2013). Acemoglu et al. (2013) find that
less constrained chiefs (chiefs in villages with fewer ‘ruling families’ 8) lead to worse
development outcomes- lower rates of education attainment, child health, non-agricultural
employment and asset ownership. Their expalnation for this finding is that chiefs, who face less
competition for their position, are freer to distort incentives to engage in economically
undesirable activities through their control of land, taxation, regulation and the judicial system
(even though their study was not conducted to investigate the effects of corruption, it gives
some insights into the issue assuming that less constrained chiefs are more likely to be corrupt).
Expenditure-tracking surveys recently undertaken by the World Bank also indicate that
corruption is costly to development(See Reinikka and Svensson, 2004; Olken, 2006;2007). For
example, Reinikka and Svensson (2004), using panel data from a unique survey of primary
schools assessed the extent to which a capitation grant to cover schools’ nonwage expenditures
actually reached the intended end-user (schools). The survey data reveal that during 1991–
1995, the schools, on average, received only 13 percent of the grants. Most schools received
8
Ruling families are families of high social status who can provide chiefs. Chiefs have to be obtained only from
these families (see Acemoglu et al. , 2013; Beekman et al., 2013;2014)
25
nothing. The largest of the school grant was captured by local officials (and politicians). They
also find that schools in better-off communities managed to receive a higher share of part of the
grant allocated to them. This implies that actual education spending, in contrast to budget
allocations, is regressive suggesting that also increases inequality (Reinikka and Svensson,
2004). Subsequent studies from other countries in east Africa have shown similar results ( be it
slightly smaller) in terms of substantial misuse of funds (Aidt, 2009). Olken (2006) in his study
of corruption in a large transfer programme of subsidized rice in Indonesia finds that, on
average, at least 18% of the rice appears to have not reached the intended beneficiaries,
suggesting that corruption may hinder developing countries’ redistributive efforts, and may
help explain the low level of transfer programs in these countries. Olken (2007) went one step
further in a related exercise in the case of corruption in rural road projects in Indonesia. He
compares the official amount reported to have been spent on the road to what he estimated what
actually the road would cost. For his estimate, engineers dug core samples of the roads to
estimate quantities of materials used, did price surveys to estimate local prices, and interviewed
households to estimate actual wages paid. Importantly, because some amount of materials
naturally disappears during construction (for example, some sand may be blown away by the
wind while on truck), Olken built several small test roads where he knew there was no
corruption so that he could calibrate the metric so it would show zero corruption when, in fact,
corruption was zero. He used the “missing expenditures” , the difference between what the
village claimed the road cost and what he estimated it actually cost, as a proxy for corruption.
Using this technique, he estimated that on average approximately 24% of the total cost of the
road was lost to corruption (Olken, 2007).
On the other hand, a (relatively small)body of literature also indicates that corrupt leaders may
have more authoritative power and may be better in organizing their communities and
resources (may lead to higher social capital) towards achieving certain development outcomes
than non-corrupt leaders. For example, Baldwin and Mvukiyehe (2011) in their study in
Liberia find that while participatory selection of chiefs increases consultation at the community
level and raises levels of participation, it also reduces levels of contributions to local public
goods. This suggests that chiefs selected by community members may be less effective at
enforcing cooperation as compared to chiefs appointed in traditional ways (possibly because
long established traditional leaders have more coercive power). Even if the above study was not
conducted in the context of investigating the impacts of corruption of village chiefs, it suggests
(assuming appointed chiefs are more likely to be corrupt than chiefs elected by the community)
26
a plausible explanation as to why corrupt leaders may be more effective in mobilizing their
community than non-corrupt leaders. In what might be perhaps more related to our study,
Turley et al. (2013) in their study of elite capture and management performance in Sierra
Leone, find that local elites are better managers of development projects, provide more
(perceived) benefits for the villagers and they find little evidence of elite capture of project
resources. Another study by Humphreys et al. (2013) similarly does not support the view of
widespread diversion of aid money at the local level in eastern Congo.
2.5.
A Historical Overview of Leadership and Corruption in Liberia
Liberia is a small country on the coast of West Africa, bordered by Côte d'Ivoire, Guinea, and
Sierra Leone. It is one of the poorest countries in the world (ranked 175 out of 187 countries on
the Human Development Index (HDI)) (UNDP, 2014). It had a total population of 4.397
million and a GDP(at current US$) of $2.027 billion in 2014 (World Bank, 2015). The country
has a predominantly agrarian economy and it is endowed with natural resources including
rubber, diamond, mineral resources and iron ore. Oil was also discovered in the country in
2012. Low human capital, low infrastructure, high unemployment and corruption have hindered
the development of the country. It had also experienced the evils of two long civil wars from
1989-1997 and 1999-2003. More recently, it has suffered significantly from the 2014 Ebola
virus outbreak in West Africa, which claimed the lives of 4,808 Liberians (The Economist,
2015 August 27). Liberia was established in 1820 by the USA as a home for freed AfricanAmerican slaves. By the end of the nineteenth century, chiefs from selected leading families of
higher social status were appointed by the Americo-Liberian settlers in a system of indirect rule
(Beekman and Bulte, 2015). Until 1980, the Americo-Liberian settlers used to lead the country
using this indirect rule system. In 1980, Master Sergeant Samuel Doe overthrew the elected
government of William R. Tolbert, becoming the first Liberian President of non-AmericoLiberian descent. In December 1989, former government minister Charles Taylor moved into
the country from neighboring Côte d'Ivoire to start an uprising to overthrow the Doe
government and this started the first civil war.
The lowest tier of leadership in Liberia is held by (town) chiefs (Baldwin & Mvukiyehe, 2011).
Historically, the process of selecting chiefs differed in different communities. As Murphy
(1980, 1981) states: “among the Mande-speaking people of western Liberia, social structure
27
was more hierarchical; the position of chief was hereditary, and only members of the
chiefdom’s founding lineage were eligible to be selected as chief. In contrast, among the Kwaspeaking people of eastern Liberia, the political system was more decentralized, and leadership
positions were more fluid” (as cited in Baldwin & Mvukiyehe, 2011). In the mid-1980s, the
Samuel Doe government introduced nation-wide elections for chiefs. The 1986 constitution
provided that registered voters shall elect paramount, clan and town Chiefs in their respective
localities and sets the term of these chiefs to be six years. In the nation-wide elections held in
the same year, some communities resisted using elections to select their chiefs (Baldwin &
Mvukiyehe, 2011). When the first civil war started in 1989, many chiefs fled or were killed,
and in rebel-occupied areas, new clan chiefs were either selected by communities or appointed
by rebel-leaders to get greater social control over rural communities (Baldwin & Mvukiyehe,
2011).
After the end of the Liberian civil war, clan chiefs were chosen through different methods such
as through participatory processes involving public votes or some other form of elections and
appointment
by higher level traditional or political leaders (Ibid.). Even though many
communities chose using public voting to select their leaders, in practice local leaders are often
still selected from among certain local elites (Beekman & Bulte, 2015). Not everybody is
qualified as potential chief and chiefs usually have to come from certain families of an upper
stratum of society (Richards et al., 2005). Richards et al. (2005) explain the leadership structure
in Liberia as follows:
Liberia is administratively divided into 15 Counties, in which the main representative of government is the County
Superintendent. S/he and her/his officials engage with a set of chiefs, whose positions are a kind of hybrid between
civil servant and constitutional monarch. The main figure is the Paramount Chief, a salaried official who mediates
between government and people, on behalf of several district chiefdoms, termed clans (though in Liberia this is
purely a territorial designation). Each clan is ruled by a district chief. Under the clan chief there are various town
chiefs and quarter elders. Most chiefdoms and clans owe something to tribal custom...................... Chiefs are
“elected”, i.e., their appointment is approved by government, after being nominated by elders. The main task of
chiefs is to preserve order, hear disputes, collect tax, recruit labour, and maintain paths and access roads, while
also conveying orders from County Superintendents and other government officials. The Paramount Chief is
advised by a council of district chiefs. The Paramount and District Chiefs are generally quite remote. Villagers
generally feel that they are more in touch with Town Chiefs than his/her superiors. Quarter elders hear the lowest
level of disputes. Some are designated as Justices of the Peace.
Liberian society is hierarchical, and studies how that many chiefs misuse their position for
private benefit (Reno, 2008; Richards and Bah, 2005 as cited in Beekman et al., 2014). Chiefs
have a big place in the local community and have various ways of extracting surplus from
28
villagers. These may occur through their power to control and allocate communal resources,
such as land and state resources channelled down from the central government, and revenues
from local enterprises (such as communal plantations). Besides, chiefs may extract surplus
from their community members directly by imposing informal taxes and through manipulated
court cases, resulting in forced and free labour on the land of powerful men (Beekman et al.,
2014).
Corruption is rampant in Liberia and the country consistently scores below 50 in the Corruption
Perception Index on a scale of 0 (highly corrupt) to 100 (very clean) (37 in 2014, 38 in 2013
and 41 in 2012) (Transparency International, 2014). Years of conflict had undermined domestic
governance structures and the rule of law, leaving the country in a state of economic chaos,
destroyed infrastructure, dysfunctional institutions, widespread corruption and unemployment.
Since the early days of coming to power, the current president Ellen Johnson-Sirleaf promised
a zero tolerance policy for fighting corruption. Acting on her promise, she fired all transitional
political appointees in the finance ministry pending an investigation into corruption allegations,
and announced that corrupt officials would be prosecuted. 17,000 government workers were
dismissed in the first months of her rule and the government then directed its attention to
longer term legal and institutional reforms (Boucher et al, 2007; Clark, 2008). The current
government has shown strong leadership in the fight against corruption which is exemplified by
its commitment to ensuring the independence of the General Auditing Commission, supporting
the establishment of the Liberia Anti-Corruption Commission (LACC), promoting transparent
financial management, public procurement and budget processes and assuring Liberia's
compliance with the Extractive Industries Transparency Initiative (EITI) through the Liberian
EITI law.
The Liberia Anti-Corruption Commission was established in 2008 with the task of investigating
corruption cases. Evidence is then provided to the Ministry of Justice for Prosecution, which is
seen by many as a major drawback. However, the LACC is empowered to prosecute any case
that the Ministry of Justice declines to prosecute, even though this has not happened in practice
(US Department of State, 2010). Low capacity and shortage of human, material and financial
resources are the major obstacles that hinder the commission to effectively fulfil its tasks. In
2009 and 2010, Executive Order 22 was issued aimed at protecting whistle blowers. However,
many public servants did not know about the order and it was not fully utilized. And civil
29
servants who report cases of corruption, graft, abuse of power, or abuse of resources are not
sufficiently protected. Some sources indicate that people who report corruption have been fired,
marginalized and persecuted. According to Executive Order 38, all government officials are
required to disclose their assets. The asset disclosure forms are kept with the Liberia AntiCorruption Commission (LACC), which has developed a guideline related to disclosure and
verification. These efforts have been pro-actively supported by the international community
and civil society through the Governance and Economic Management Assistance Program
(GEMAP). These combined efforts have led to a remarkable progress in terms of curbing
corruption in the country. While Liberia still scores below averages in many aspects of
governance as compared to other countries, it has shown positive changes in most indicators of
governance since President Johnson-Sirleaf took office in 2006 (Transparency International,
2012).
However, in spite of these positive developments, corruption remains endemic and exists in
most sectors of the society. Low public sector salaries, lack of training and capacity, inefficient
and cumbersome regulations are considered the drivers of corruption across the public sector.
In addition, Liberia is endowed with vast mineral wealth including iron ore, timber, diamonds,
gold and rubber, and in spite of progress made in this area too, natural resource management
continues to face major corruption and governance challenges (Transparency International,
2012). Citizens interviewed within the framework of the 2008 and 2012 Afro barometer9
survey also perceive corruption to largely exist in most national institutions. The police and
revenue collection were perceived to be the most corrupt institutions in the 2008 survey, with
respectively 52 % and 46 % of the respondents believing that all or most of the police and tax
officials to be involved in corruption, followed by national government officials (45%) and
parliamentarians (44%) (Afrobarometer, 2008). Interestingly, public perception about
corruption amongst most or all who are tax officials remains the same, since 2008
(Afrobarometer, 2013). Corruption continues to be a hindrance to the provision of public safety
in Liberia. A majority of the respondents in the 2012 survey (55 percent) believes most or all of
the police are corrupt. Overtime, strong perception about police corruption remains consistent
with three percentage points increase from 52 percent in 2008 to 55 percent in 2012 amongst
the respondents who believe that most or all in the police are corrupt. Overtime, negative
9
The Afrobarometer is an independent, non-partisan research instrument that measures the social, political
and economic atmosphere in selected African Countries. It is a comparative series of public opinion surveys
that measure public attitudes toward democracy, governance, the economy, leadership, identity, and other
related issues. In 2008, the Afrobarometer surveyed a nationally representative sample of 1200 adult Liberians.
(Afrobarometer, 2008).
30
perception about corruption in the presidency decreased by 10 percentage points from 56
percent in 2008 to 46 percent in 2012 amongst those who believe corruption is only amongst
some officials in the presidency. However, a sizeable proportion (31 percent) of the population
between 2008 and 2012 have consistently perceive that most or all officials in the presidency
office are corrupt. A large section of the public perceives elected officials in the legislature to
be corrupt. Thirty eight percent perceive most or all members of the legislature to be corrupt in
the 2012 survey. The segment of the population that perceive most or all national officials as
corrupt increased by six percentage points from 40 percent in 2008 to 46 percent in the 2012
survey. The percentage of respondents who perceive most or all who are judges and magistrates
are corrupt increased from 37 percent in 2008 to 43 percent in 2012 (Afrobarometer, 2013). A
large portion (32 %) of the respondents in the 2008 survey believes that traditional leaders
(which include town chiefs) are corrupt (Afrobarometer, 2008). Unfortunately, popular
assessment of corruption among traditional leaders was not included in the 2012 survey and
hence we are not able to see the changes in perception overtime regarding the corruption of
traditional leaders or chiefs.
31
Chapter Three: Data, Model specification and Estimation Strategy
In this chapter, we explain the data source, describe the data we used, elaborate the estimation
techniques we employed to achieve the objective of the study and set out our hypotheses.
First, we provide an overview of the data used and data source. Then, we specify the
econometric models and explain the estimation strategy used to answer the research questions.
Following that, we define the dependent and independent variables of our models. Finally, we
set out our hypotheses and justify our expectations.
3.1. Data Source and Study Context
This study is based on a novel micro dataset collected from 132 randomly selected villages in
rural Liberia in 2012 and 2014. The data were collected in the context of a large scale impact
evaluation study for the Dutch Ministry of Foreign Affairs under its MFS Programme. MFS II
is the grant framework for Dutch NGOs (Co-Financing Agencies, CFAs) with the main aim of
directly contributing to poverty reduction in developing countries. The baseline data for the
evaluation study were collected during May-August 2012. Additional baseline data were
collected in November 2013. The follow up survey was conducted during May-June 2014
(Beekman et al., 2015). The corruption of the chiefs was measured in 2012 before the baseline
study. Carefully measured bags containing 30 kgs of rice seeds (intended for beneficiaries of
the one of the projects in the villages) were kept at the chief’s house for 2 days in each village
and then the bags were re-measured to see if any amount was missing. So, the percentage of
missing rice seeds (‘missing input %j’) is our proxy for the corruption of the chiefs. To check
the robustness of our results, we also create a binary variable (‘missing seed’) indicating
whether any rice seed was missing (Beekman et al., 2013). The other demographic and socioeconomic data regarding both the village chiefs and households were collected as part of the
impact evaluation study using standardized questionnaires administered at the household and
community levels.
3.2. Model Specification and Description of Variables
To answer the first research question, we specify a simple censored regression model to be
estimated using maximum likelihood technique to explain variation in corruption among
village chiefs. We chose a censored regression model because we have a significant number
of observations indicating zero missing rice seeds (almost half of the chiefs in our sample did
not ‘steal’ the rice seeds) and hence we have a censored data. Using OLS in this case leads to
biased and inconsistent estimates. Our model specifies corruption of the chief as a function of
32
the chief’s characteristics, community characteristics, other socio-economic and cultural
factors that can potentially affect corruption. Our model is mathematically specified as:
Missing input (%) j = α + γk +β1Commj + β2Chiefj + β3Z + εj ....................(1)
Where subscript j denotes community j, subscript k denotes district k, α= constant, γk =
district fixed effects to capture unobservable factors that vary at district level, Comm j =
community controls, Chiefj = chief’s characteristics and Zj = other relevant variables expected
to influence corruption behaviour. Our main parameters of interest in the model above are β2
& β3
The dependent variable (‘missing input %j’) in our study is the percentage of rice seed
missing (after the seeds have been kept at the chief’s place for 2 days and then measured)
which is a continuous variable used as a proxy for the corruption of the chief. Our main
explanatory variables (‘Chief’) include the chief’s characteristics such as: the age, gender,
ethnic origin, education level, economic position, household size, private landholding, term in
office and religion. Other covariates (Z) include: average literacy rate in the village, the
number of ruling families in the village, ethnic fragmentation of the village, number of
households with mobile phones, TV, radio in the village, frequency of visits of the village by
officials, villagers perception of the likelihood of corruption of the chief, distance of the
village to the nearest major town, frequency of criticisms of the behaviour of the chief by the
community during meetings, reaction of the community when the chief misbehaves, whether
new chiefs are obtained through election or are appointed, whether corruption is defined
positively in the village and the amount of resources that villagers believe the chief can
confiscate as tax.
Our community controls (Commj) include gender composition in the village (to control for
risk preferences), previous NGO presence, the share of young population in the community
(to control for risk taking preferences), size of communal rubber plantations (to control for
different socio-economic and labour market conditions), village size (number of households
in the village), household attacks/ exposure to violence (to control social and risk preferences)
and the share of displaced people in the community( to control social/ risk preferences and
corruption). Binary variables indicating whether the village has access to the main road
(market integration), access to the local market, whether some village members were recruited
to war and the availability of off-farm work will also be included (to control for different
levels of socio-economic and institutional quality) .
33
It is important to control gender composition in the village because many studies indicate that
women are more risk averse than men. So, communities may vary in terms of their risk
preferences due to differences in gender composition and this variation in risk preference
may affect the corruption behaviour of the chief, say for example through monitoring.
Controlling for NGO presence is relevant as projects may influence both corruption and social
or risk preferences. Previous NGO projects might have invested in educating the community
and leaders on the adverse effects of corruption and ways of fighting against it. Studies
indicate that young men show more risk-taking behaviour than other people, and this may
influence social preferences in the community. It is important to control socio-economic and
labour market conditions that may affect corruption. This is captured by the size of communal
rubber plantations. Village size, exposure to violence and share of displaced people are also
controlled as these may affect social risk preferences and corruption. For example, studies
indicate that exposure to violence affects trust (though the results on the sign are mixed)
among villagers in the post-conflict era (see Richards et al., 2005; Voors et al., 2012) and this
may in turn affect corruption. It is also necessary to control for differences in market
integration as these may affect corruption say, for example, through the diffusion of
information. Differences in socio-economic and institutional quality may be related to
corruption. For example, villages with poor institutions may be expected to have more corrupt
chiefs. These differences are captured by controlling access to local market, whether some
members were recruited to war and availability of off-farm work.
In addition we specify and estimate the following interaction model to answer the second
research question:
Missing input (%) j = α + γk +β1Commj + β2Chiefj + β3 Z + β4 Chiefj * Z+ εj...... (2)
Where β4 captures the conditioning effect of the covariate variables on corruption of the
chief.
Equations 1 & 2 may suffer from endogeneity bias due to reverse causality ( for example, due
to the fact that more corrupt chiefs may be monitored more frequently) and/ or due to other
factors not included in the model that drive both some of the explanatory variables, say for
example, the average literacy rate in the village and corruption of the chief. Hence, it may be
difficult to establish strict causal relationships between the explanatory variables and the
corruption of the chief. So, our results should interpreted taking this limitation into account.
But many studies also show that the causality most likely goes in one direction- from the
34
characteristics of the chief to the corruption behaviour of the chief, and not the other way
round. For example, Richards et al. (2005) explain that villagers in rural Liberia seldom
remove misbehaving chiefs from power. We also argue that many of the chief’s
characteristics (for example, private land holding) are exogenous in the Liberian context (See
Richards, 2005; Beekman et al., 2014).
Table 2: Description of Variables
Variable category
Variable
Description
Dependent Variable
Missing input (%)
Percentage of rice seed missing
Missing Seed (b)
If any amount of rice seed is missing (1=Yes)
Chief Characteristics
Age
Age in years
(‘Chief’)
Gender
Gender of chief (1=male)
Ethnic origin
Ethnicity of chief (1=major tribe)
Education
Highest grade completed in school (years)
Economic position
Number of meals per day
Household size
Number of people eating from the same pot
Landholding
Acres of land privately owned
Term
Number of years in office
Other covariates (Z)
Community
(Comm)
controls
Religion (b)
Religious affiliation of the chief (1=Protestant)
Born (b)
Whether chief was born in the village (1=Yes)
Off-farm (b)
Whether the chief is engaged in off-farm job (1=Yes)
Elected (b)
Whether chief is elected (1) or appointed
Literacy
Share of literate people in the village
Ruling families
Number of families who can provide chiefs
Ethnic fragmentation
Herfindahl Index
Mobile phones
Number of mobile phones in the village
Visits
Number of Visits by officials
Perception of corruption
Villagers’ perception of chief’s corruption
Distance
Distance to major town in hours
Criticisms
Frequency of criticisms by villagers
Reaction (b)
Number of reaction of villagers when chief misbehaves
Corr-distr (b)
Whether the district commissioner is (perceived) corrupt
Definition (b)
Whether corruption is defined positively
Tax entitlement
Amount of resources chief is entitled to confiscate as tax
Gender comp
Share of women in the village
NGO (b)
Previous NGO presence (1=Yes)
Share of Youth
Share of young people in village (12-25 years)
Village size
Number of households in village
Rubber plantations
Acres of plantations in village
Violence (b)
Number of household attacks
Market (b)
Availability of market in village (1=Yes)
(b) indicates the variable is binary.
35
3.3.
Hypotheses and Expected Signs
Based on our theoretical framework and the findings from previous empirical studies on
determinants of corruption, we propose potential determinants of the corrupt behaviour of
chiefs to be investigated in our study. Table 3 below provides these potential variables along
with their expected signs.
Table 3. Proposed Variables and their expected signs
Variable or factor
Expected sign
Age
+
Male chiefs
+
Ethnic origin
+
Education level of the chief
+/-
Economic position of the chief
-
Household size of the chief
+/-
Size of the private land holding of the chief
+
Chief’s term in office
+/-
Protestant Chief
-
Whether the chief is born in the same village
+
Whether the chief has offfarm job
-
Number of ruling families
-
Reaction of the community when the chief misbehaves
-
Distance of the village to the nearest major town
+
Number of mobile phones in village
-
Ethnic heterogeneity/ fragmentation of the village
+
Elected chiefs
-
Chief’s entitlement to confiscate a tax
+
Positive definition of Corruption
+
Criticisms by the community
-
Visits by government officials
-
Perception of chief’s corruption
+
Perceived corruption of district commissioner (Corr-distr )
+
Below we explain our justifications for the expected signs of the variables or factors we
proposed in Table 3.
We would expect older chiefs to be more corrupt than younger chiefs. A possible theoretical
explanation might be older chiefs may have more experience in hiding their corrupt activities
and have identified credible partners in corruption overtime. Another explanation might be
36
older chiefs have shorter time horizon and hence their expected costs if caught in corruption
might be lower. On other hand, younger chiefs may have longer time horizon and may be less
likely to engage in corruption as their expected costs might be higher. It might be argued that
younger chiefs may be more risk-taking and may opt to engage in corrupt practices but we
believe that this risk preference is likely to be offset by their long-term time horizon.
Many experimental studies indicate that males tend to be more corrupt than females (see
Chaudhuri, 2012). In fact, according to Chaudhuri (2012), no studies exist that find men to be
less corrupt; either studies show that females are less likely to be corrupt or there are no
significant gender differences in relation to corruption. Possible explanations include that
females are more pro-social and risk-averse and also less committed to reciprocity. Therefore,
we would expect male chiefs to be more corrupt than female chiefs.
We would expect chiefs from the major ethnic group to be more corrupt. For example,
Beekman et al. (2014) in their study in rural Liberia find that chiefs belonging to the major
ethnic group tend to be more corrupt. Their possible explanations for this include that chiefs
from the major ethnic group may be less scrutinized and less likely to be punished by the
community and/or may have higher need for stealing to channel resources through ethnic
patron-client networks. Chiefs may divert community resources and distribute them among
their co-ethnics in order to get support and approval (Beekman and Bulte, 2015). On the other
hand, it can be argued that the flow of cheap information through the community and the
consequent social sanctions if corruption is discovered can lead to lower corruption if the chief
is from the major ethnic group. But rural societies in Africa are mainly characterized by their
loyalty to non-state institutions like their ethnicity which implies that such a flow of
information puts more pressure on the chief to please co-ethnics even by acting corruptly since
this is more expected and favoured than being loyal to one’s position.
We would expect more educated chiefs to be more responsible, honest and prosocial leaders
due to the expected role of education in raising ones moral standard and understanding of the
(long-term) consequences of one’s actions. On the other hand, it might be the case that more
educated leaders are expected to be able to easily manipulate their community through various
means and easily cover their actions from higher officials and villagers. Besides, earlier studies
also indicate that more educated people have lower risk aversion and in our specific case more
educated chiefs may underestimate the risk of detection and punishment and hence may be
37
more inclined to engage in corrupt activities. Hence we expect that the relation between the
education level of the chief and corruption behaviour can go in either direction.
We would expect chiefs with better economic positions to be less corrupt. Possible
explanations might be lower need for additional income (or stealing resources) and higher
feeling of shame and guilt or dishonesty cost (if detected) because of already higher social
status. A related argument might be if the chief’s economic position is worse, his or her corrupt
practices might be tolerated by the community and thus reducing the possibility of sanctions
and associated costs. Many cross-national comparison studies of determinants of corruption
have found that corruption is more common and severe in countries with lower public sector
wages and the majority of lab experiments on anticorruption policies find that paying fair
salaries to officials decreases corruption (see Abbink and Serra, 2012). We will not be able to
test this directly in our study since the position of the chiefs is unpaid in rural Liberia and as a
result we do not have data on chief’s wages. But still we believe that similar reasoning can be
used to test the effect of the economic position of the chief on his/her corruption behaviour. In
fact, since the chief is unpaid, some stealing of public resources may considered a
compensation for his/her services especially if his/her economic position is worse.
The household size of the chief is expected to affect the chief’s corruption behaviour. An
increase in the chief’s household size (defined as the number of people eating from the same
pot) is expected to increase the consumption and other expenditure needs of the household,
other things remaining equal. As such, with an increase in the chief’s household size, we expect
an increase in the need for more income or resources and hence a greater pressure to steal
community resources (especially with a larger share of dependent members in the household).
On the other hand, if the household consists of more working or income earning members than
dependents, it may also be the case that the need for stealing community resources may be
lower. It is also possible that household size may be influenced by another factor, say, income
that may affect the corruption behaviour of the chief. It may be the case that better off chiefs
have a larger household size and have less need to steal and vice-versa.
The size of private land holding of the chief is expected to increase the marginal benefit of the
missing inputs by raising their marginal value product and/or lower the expected cost of
stealing the inputs by lowering the probability of being detected since it will be easier to hide
the use of stolen inputs in a better way on large plots. In either case, large land holdings are
expected to derive corrupt behaviour. Beekman et al. (2014) in their study in rural Liberia find
38
that chiefs with larger private landholdings are more corrupt. On the other hand, chiefs with
large holdings may be assumed to have a better economic status and income and hence have
lesser need to steal the inputs than chiefs with small land holdings. For our specific case, we
would expect a positive relation between the size of private land holdings of the chief and
corruption behaviour due to the nature of the item we used to proxy corruption (rice seed). The
rice seed is of a specific type which can be easily recognized by villagers as it grows and hence
chiefs with large landholdings can relatively easily hide this type of rice from villagers.
The chief’s term in office may be expected to be related to the corruption behaviour of the
chief. On the one hand, longer term serving chiefs may be expected to identify and develop
corrupt partners overtime and/or have more experience in hiding their corrupt activities and
hence may be more corrupt. On the other hand, longer term serving chiefs might have to
behave properly and be less corrupt to stay in power.
We would expect chiefs belonging to the Protestant religion to be less corrupt. Many empirical
studies have shown that countries with protestant traditions are perceived to be less corrupt
(Treisman, 2000; Lambsdorff, 2005). Possible explanations for this include: a greater tolerance
for challenges to authority and for individual dissent makes Protestant societies more likely to
discover and punish official abuses; Protestant cultures are less understanding toward
deviations from virtues and press more urgently to institutionalize virtue and eliminate the
wicked. In addition, ‘a focus on the family rather than the individual in many traditions other
than Protestantism leads to ‘amoral familism’ (favouring the economic interest of the nuclear
family over the public interest) and nepotism; and Protestant traditions-in which the separation
of church and state is more pronounced than in, say, Catholicism or Islam- lead to a more
vibrant, autonomous civil society that monitors the state more effectively’ (Treisman, 2000, P.
428).
Chiefs who were born in the village may be expected to have more family and friends network
than chiefs who were born elsewhere and hence this may encourage the misuse of power to
benefit members of the network. Besides, such chiefs may face less scrutiny and sanctions from
villagers as they are considered one of their own and thus face lower expected cost of engaging
in corruption. Both these factors may reinforce corrupt behaviour by the chief. Hence, we
would expect chiefs who were born in the village to be corrupt as opposed to chiefs who were
not.
39
The number of families who could provide chiefs, or ruling families as they are called, is
regarded as an indicator of the presence and degree of competition for the position of the
incumbent chief. For example, Acemoglu et al. (2013) in their study of the effect of constraints
on chiefs’ power in Sierra Leone hypothesize and show that the greater the number of ruling
families in a chieftaincy, the greater the extent of political competition. This in turn implies
more constraints placed on the power of a ruling chief. ‘Even if one family is able to dominate
the chieftaincy for many generations, with more ruling families, there will be a greater potential
to lose the paramount chieftaincy in an election- creating a powerful threat that will discipline
paramount chiefs, forcing them to govern better’ (Acemoglu et al, 2013, p. 2). The empirical
findings (at macro level) on the relation between democratic competition and corruption is
mixed. Many cross-national studies show that (long-term)democratic countries are less corrupt
(for example, see Treisman, 2000; Lambsdorff, 2005). This can be explained in terms of the
role of access to information, free press and the independence of the legal and electoral systems
in a democratic society which will create a disciplining mechanism by increasing the expected
costs of engaging in corrupt behaviour. On the other hand, some of these studies have found no
effect or a negative effect of (short-term) democracy on corruption suggesting that only high
level and long-term democracies are able to curb corruption. For example, Treisman (2000)
finds that current level of democracy does not explain corruption while Lambsdorff (2005)
finds that medium democracies are related with high corruption (see also Manow, 2005 cited in
Lambsdorff, 2005). It might also be theoretically argued that with increased political
competition, the threat of losing power may induce the incumbent to try to grab as much and as
quickly as possible before potentially losing power and thus leading to more corruption. But we
expect that for such a situation to occur, the benefits from the graft should be far greater than
the long term benefits of staying in power (behaving properly) as well as the costs (including
social and psychological) of losing the paramount chieftaincy upon detection of corrupt
behaviour. For our specific case, we think this is less likely to be the case mainly because of the
low value and non-durable nature of the resources we used to proxy corruption and hence we
would expect a negative relation between the number of ruling families and corruption of the
incumbent chief.
The more people criticize a chief, the more visits of village leaders by officials and the more
the reaction of the community when the chief misbehaves can be regarded as indicators of the
level of monitoring and (social) sanctions. As such, following our theoretical framework, these
are expected to reduce corrupt behaviour by increasing the probability of detection and its
40
expected costs. But, on the other hand, it can also be argued that increased monitoring may
crowd out intrinsic motivations for honesty and reduce trust and hence lead to more corrupt
behaviour (See Armantier and Boly, 2011; Soreinde, 2014). For our case, we would expect a
negative relation of these factors with corrupt behaviour of the chief.
The average literacy rate in a village can be regarded as an indicator of the ability of the village
community to obtain and process available information and to differentiate their rights and
duties as well as of their leaders. More educated people are expected to have better information
regarding the budget allocated to the village and can discipline the chief by strict monitoring.
They are also expected to be well-informed about alternative ways of exposing corrupt
practices once discovered and hence raising the threat of punishment of engaging in corrupt
practices (can easily communicate the media, higher level officials etc.). In addition, they are
expected to be able to keep records to hold their leaders accountable and which can be used as
evidences in the prosecution of corrupt leaders. Expansion of education makes hiding abuses
and crimes difficult (Treisman, 2000). So, we would expect a negative relation between the
average literacy rate in a village and corruption of the village chief.
Most studies have documented that perception of corruption is correlated with actual
corruption. More (perceived) corruption makes it easier to find a corrupt partner (Adving and
Moene, 1990). In addition, a higher estimated level of corruption may reduce an individual’s
moral costs of committing a crime (may lead to lower dishonesty cost, d following our
theoretical framework), if corrupt practices are the norm rather than the exception. As indicated
in the social learning theory in our theoretical framework, a willingness to engage in corruption
may increase if there is a perception that such a behaviour is widespread and, thus, approved
(Tavits, 2005). Another possible explanation might be our perception influences our action and
hence people may tolerate actual corruption when their perception of corruption is high. Or,
conversely, if there is trust or low perception of the likelihood of corruption, it encourages
positive reciprocity and compliance leading to lower corruption (trust leads to social bonds).
So, we would expect a positive relation between the villagers’ perception of the likelihood of
corruption of the chief and the actual corruption behaviour of the chief.
In ethnically divided societies, villagers may provide cheap information about and even put
internal sanctions against those who betray their coethnics (Fearon and Latin, 1996). This may
foster nepotism and clientelism. Some studies also indicate that people are more likely to
engage in corruption if there is quite some distance between them and the ones affected as a
41
result of the corrupt activities. Beekman et al. (2014) find that chiefs in ethnically
heterogeneous communities in rural Liberia are more likely to be corrupt. Beekman and Bulte
(2015) refine the results from the earlier study by Beekman et al.(2014) and find that corrupt
chiefs target their communities along ethnic lines where the predation or corruption by chiefs
negatively affects non-coethnics more than their coethnics. So, we would expect positive
relation between ethnic fragmentation in a village and corruption of the chief.
We would expect distance of the village to the nearest major town to be positively related to the
corruption of the village chief. This may be due to less communication possibility to gain more
information for monitoring and punishment (lower access to media, higher or regional
government officials etc. to report corrupt activities) or the inaccessibility of the village for
regular monitoring, which will lower the chance of detection by higher officials. In the same
manner, we would expect the availability of different communication and media facilities to be
related with lower corruption. Accordingly, the larger the number of people with mobile
phones, TVs and radios, the lower the corruption of the chief, other things remaining constant.
Information access and transmission enhance monitoring and punishment and thus reduce the
likelihood of corruption. To quote Winters et al. (2012, p.216),
.... corruption is a result of asymmetric information. Agents (either elected or bureaucratic
officials) possess information that their principals (either voters or the general public) do not, and it is this lack
of information that creates the opportunity for corruption to occur. Agents weigh the benefits that can be gained
from corruption against the expected costs of getting caught, which are a function of the availability of
information and the likelihood of punishment. Holding the punishment mechanism constant, when information is
scarce, the probability that a corrupt act will be discovered is low, and so the expected cost of corruption is low,
resulting in more corruption. Similarly, holding information constant, when the punishment is weak, the expected
cost of corruption is also low, resulting in more corruption. If there is little information that might reveal
corruption or else little punishment when corruption is exposed, then agents will act with impunity.
In addition, the media and information play a great role in educating the community about their
rights and responsibilities thereby enabling them to scrutinize the actions of the chiefs and hold
them accountable. So, we would expect a negative relation of corruption of the chief with the
number of people having mobile phones.
If a new chief is obtained through election, we would expect the incumbent chief to be less
corrupt. Election entails the risk of being removed from power if one misbehaves. So, elected
chiefs are expected to be less corrupt and more accountable to the electorate than appointed
chiefs due to fear of punishment of being removed from their position in case of inappropriate
behaviour. On the other hand, appointed chiefs tend to be more corrupt due to absence of
42
competition or less fear of being removed from power and are thus expected to show less
accountability.
The more corruption is defined favourably or positively, the higher the likelihood of the
corruption of the chief. This follows from the social learning theory explained in our theoretical
framework and from empirical findings. As stated before, a willingness to engage in corrupt
behaviour is expected to increase when one does not define corruption as morally or
situationally wrong, but rather considers it an acceptable mode for exchange (Tavits, 2005). In
such a situation, there will be lower dishonesty cost (lower guilt feeling or shame) due to the
social acceptance of ‘corruption’. Moreover, the community will be less willing to monitor and
put social sanctions on corrupt practices since corruption is a tolerated norm (for example,
bribe giving is considered conforming to social reciprocity in some societies). The same
reasoning applies if villagers believe that chief is entitled to confiscate a tax. The more the
amount of community resources that the villagers believe the chief can confiscate as tax, the
more the corruption of the chief will be tolerated and thus the more the incentive for corrupt
behaviour.
43
Chapter Four: Results and Discussions
In this chapter, we present & discuss the results of our analysis. We first provide the summary
statistics of chief & village characteristics. We then present & interpret the results of the
econometric estimations we used to test our hypotheses. As stated in the previous chapter, we
estimated two equations. The first equation relates the corruption of chief with the chief’s
characteristics & other covariates of corruption behaviour. The second equation considers the
interaction effects of some of the chief & village characteristics on the corruption behaviour
of the chief. Starting from the most parsimonious model, we present & discuss our
estimations. Finally, we discuss the assumptions & limitations of our analysis & highlight the
caveats that should be considered in interpreting our results.
4.1 Summary Statistics of Chief & Village Characteristics
Table 4.1 presents the summary statistics of chief & village characteristics.
Table 4.1 Summary statistics of variables
Variable
Mean
SD
Min
Max
N
Missing input (%)
3.10646
5.253004
0
28.66667
129
Missing
.931938
1.575901
0
8.6
129
Missing seed(b)
.5503876
.499394
0
1
129
Age
48.176
11.3223
27
82
125
Gender(b)
.8629032
.3453448
0
1
124
Ethnic origin(b)
.6870229
.4654852
0
1
131
Education
5.390244
4.144192
0
14
123
Economic position
1.6
.5819877
0
3
125
Household size
11.38889
13.96566
1
68
125
Landholding
81.608
127.0404
0
600
125
Term
6.952
6.738019
0
32
125
Religion(b)
.8396947
.3682974
0
1
131
Born(b)
.6030534
.4911429
0
1
131
Off-farm(b)
.3435115
.4767033
0
1
131
Share of literate
.4118923
.1281945
.0707071
.9826087
130
Ruling families
4.152
2.876647
1
10
125
Ethnic fragmentation
.3645632
.2493513
.835
130
Mobile phones
163.0079
276.137
2
1500
127
Visits
1.381108
.1710322
1
2
135
Corruption perception(b)
.3664122
.4836736
0
1
131
Distance
22.00775
16.91638
1
90
129
Criticisms
1.406812
.1526408
1
2
135
Reaction
2.333333
2.603098
0
11
135
Elected(b)
.5877863
.4941228
0
1
131
Positive definition(b)
.0992366
.3001272
0
1
131
44
0
Tax Entitlement
Corr-distr (b)
2.504923
.6211383
1.44
4.3
130
.6717557
.4713768
0
1
131
Gender composition
.501364
.0478581
.3731343
.6185567
130
NGO(b)
.6793893
.4685029
0
1
131
Share of youth
.2704263
.0579786
.1392405
.4112903
130
Village size
163.8672
257.5716
12
1944
128
Rubber plantations
212.4264
1107.915
0
12000
129
Violence
7.152672
2.78588
0
15
131
Market(b)
.5511811
.4993434
0
1
127
(b) indicates the variable is binary
From Table 4.1, we can see that, on average, around 3 % (≈0.9 kgs) of the 30 kgs rice seeds
distributed were missing. Rice seeds were missing in around 55 percent of the villages on
average (in 72 out of 130 villages). This suggests the existence of widespread corruption
among village chiefs in rural Africa consistent with the findings of previous studies(assuming
that our measure proxies corruption behaviour in a real environment). The average age of the
chiefs is around 48 years. The majority (≈86 per cent) of the chiefs are males & around 68
percent of the chiefs belong to the major ethnic group (kpelle). On average, the chiefs have
about 5 years of schooling & a household size of 11 people. The chiefs’ household on average
have two meals per day (used as an indicator of the economic position of the chief’s
household). In terms of private land holding, the chiefs on average own around 82 acres of
land but there is significant variation among the chiefs (SD≈127). This average landholding of
the chiefs is much larger than the average land holding for villagers in rural Liberia. This may
suggest that the chiefs (mis)use their power to unfairly influence the allocation of resources
such as land to their advantage. But, given the nature of land allocation in Liberia which is
determined centrally, this is less likely to happen and we argue that landholding of the chief is
exogenous to corruption behaviour. The average chief has served the community for about 7
years. The majority of the chiefs (≈84 percent) belong to the Protestant religion & a
significant portion of the chiefs (≈60 percent) were born in the same village. Around 34
percent of the chiefs are engaged in off-farm activities. The summary statistics also show that
the share of literate people in the villages is on average 41 percent. On average about 4
families could provide chiefs (so called ‘ruling families’) in the villages. This is a bit higher
than what is indicated in previous studies. For example, Beekman & Bulte (2015) indicate
that in their studies, on average, 2 families are eligible to provide a chief in a typical village.
But, in that study, the number of villages considered is much smaller (44) than in our study
(136) and this may explain the difference. The villages are mainly characterized by less
45
heterogeneity in terms of ethnic composition, the average Herfindhal index for ethnic
heterogeneity being around 0.36.
The average number of mobile phones in the villages is 163 with a very wide variation among
the villages (with SD of about 276). On average, in about 37 percent of the villages, members
perceive that their chiefs are corrupt. It takes on average about 22 minutes from a typical
village to reach the nearest town. In a typical village, on average, 2 villagers report to higher
officials when the chief misbehaves. A significant number of the chiefs (≈59 percent) took
their position through election. Relatively only a small portion of the villages (around 10 per
cent) positively define acts of corruption (or consider them acceptable). On average, villagers
consider that their chiefs are entitled to take about 2.5 out of 10 resources directed to the
village as a tax. In a significant portion of the villages (≈67 percent), villagers perceive that
higher district officials are corrupt. This may suggest the prevalence of corruption among
higher district officials. On average, females constitute around 50 percent and the youth (aged
between 12-25) around 27 percent of the total population in the villages. These indicate a
relatively balanced gender composition in the villages and a higher share of
villagers
belonging to the non-youth category. We included these gender composition and share of
youth as controls as studies show that these variables are related to risk preferences. We also
included previous NGO presence, village size, rubber plantations, exposure to violence &
access to a nearby market as controls. A significant number of the villages (around 68
percent) have benefitted from previous NGO activity. This is also consistent with the results
of previous studies. For example, Beekman et al. (2014) report similar findings & explain that
this high percentage of NGO presence may be the result of a number of reconstruction efforts
in post-war Liberia. The villages are typically of medium size with an average number of
households of around 163. Rubber plantations are important in rural Liberia. On average,
around 212 acres of rubber plantations are communally owned in a typical village. As
indicated in the previous chapters, conflict & violence have been an important feature of
Liberian society. On average, around 7 households have been exposed to attacks in a typical
village. Around 55 percent of the villages have access to a market that can be reached within
one hour on foot .
4.2 Econometric results
In this sub-section, we provide & interpret the results of our econometric estimations starting
from the most parsimonious model.
46
4.2.1 Determinants of Corruption: Main effects
Tables 4.2.1 presents the results of the first category of our estimations. First, we estimated a
simple OLS model to identify which chief characteristics determine corruption behaviour. As
indicated in the previous chapter, OLS gives biased estimates since our dependent variable is
censored. Hence, we present its result (in column 1) for comparison & will not interpret it in
detail. Then, in column 2, we estimated a censored regression model to identify which chief
characteristics determine corruption behaviour while excluding community controls & district
fixed effects. Contrary to our expectation, the results show that only one of the chiefs
characteristics significantly determines corruption behaviour. We find a positive and
significant effect (at 5 percent) of being born in the same village on the corruption of the
chief (‘stealing’ of rice seeds). In the context of our theoretical framework, this may be
explained in two ways. First, chiefs who were born in the same village may face less scrutiny
and sanctions for their misbehaviour from villagers as they may be considered one of their
own. This will reduce the expected cost of engaging in corruption and hence reinforce corrupt
behaviour Second, chiefs who were born in the same village may have more extended family
and friends networks and hence face more pressure to divert community resources to satisfy
the demands of their networks. This may increase the expected benefit of engaging in
corruption and drive corrupt behaviour. In the next step (column 3), we estimated a censored
regression model with district fixed effects included to account for unobservable factors that
may vary at district level. In this specification also, we find a positive and significant effect
(at 10 percent) of being born in the same village on corruption behaviour. We do not find
significant effects of the other chief’s characteristics on corruption behaviour. In column 4,
we tried to identify other factors that may determine corruption behaviour besides the chiefs
characteristics. In particular, we examined which village characteristics may explain the
variation in the corruption of the chiefs. We included both district fixed effects and
community controls in this censored regression specification. Again, being born in the same
village positively & significantly (at 10 percent) affects corruption behaviour. So, this
variable robustly determines corruption behaviour in various specifications as can be seen in
Table 4.2.1. We now also find that the (perceived) corruption of the district commissioner
positively & significantly (at 10 percent) affects corruption of the chief. Within the context of
our theoretical framework (social learning theory), we label this as imitation where problem
behaviour is acquired from peers/ colleagues and /or subordinates. In villages where the
district commissioner is perceived to be corrupt, chiefs may imitate such behaviour as they
may consider it widespread and thus approved.
47
Table 4.2.1 Determinants of Corruption: Main effects
---------------------------------------------------------------------------------------------Variable
Model 1
(OLS)
Model 2
(Tobit)
Model 3
(Tobit)
Model 4
(Tobit)
-----------------------------------------------------------------------------------------------
Age
Gender
ethorigin
Education
Eco_position
-0.028
-0.023
0.001
-0.016
(0.05)
(0.07)
(0.07)
(0.07)
0.196
0.170
0.244
1.500
(1.39)
(2.21)
(2.26)
(2.31)
0.328
-0.394
(1.15)
(1.80)
(1.78)
-0.795
(1.69)
0.343
-0.143
-0.294
-0.277
-0.191
(0.12)
(0.19)
(0.19)
(0.20)
-1.095
-1.568
-1.850
0.059
(0.84)
(1.31)
(1.40)
(1.34)
HHsize
-0.050
-0.069
-0.061
-0.065
(0.03)
(0.06)
(0.06)
(0.06)
Landsize
0.010*
0.011
0.007
0.003
(0.00)
(0.01)
(0.01)
(0.01)
0.021
0.085
0.059
-0.107
term
(0.08)
(0.12)
(0.12)
(0.11)
protestant
-0.721
0.440
-0.501
-0.189
(1.41)
(2.28)
(2.26)
(2.20)
born
3.273**
5.127**
3.973*
(0.99)
(1.59)
(1.62)
offfarm
elected
shareliterate
3.355*
(1.54)
0.862
1.177
1.889
2.434
(1.01)
(1.59)
(1.61)
(1.50)
-0.699
-1.827
-1.207
(1.00)
(1.58)
(1.75)
-2.664
(1.65)
4.148
(6.67)
rulingfamilies
-0.080
(0.28)
ethnichetero
-1.484
mobilephones
0.002
(3.59)
(0.00)
visits
1.460
corr_perception
1.403
Distance
-0.007
Criticisms
-5.569
Reaction
0.228
(4.50)
(1.63)
(0.06)
(5.25)
(0.31)
positivedef
-1.930
48
(2.61)
aventitlement
1.457
Corr-distr
4.718*
(1.11)
(1.91)
constant
5.019
(3.26)
2.246
1.898
- 1.556
(17.02)
(5.01)
(5.47)
District FE
No
No
Yes
Yes
Community c.
No
No
No
Yes
----------------------------------------------------------------------------------------------sigma
constant
7.282***
6.992***
5.512***
(0.67)
(0.64)
(0.52)
----------------------------------------------------------------------------------------------R-sqr
0.173
F-statistics
1.886
Chi2
PseudoR^2
N
19.382
27.158
51.951
0.035
0.049
0.103
121
111
121
121
--------------------------------------------------------------------------------------------------
*,** & *** represent significance at 10%, 5% & 1% respectively.
Standard errors in parenthesis
4.2.2 Determinants of Corruption: Interaction effects
In our next exercise, we first estimated an interaction model to investigate if there are any
interaction effects of some of the chiefs and village characteristics on the corruption of the
chief. The results are presented in Table 4.2.2. In column 1, we estimated a censored
regression model with four interaction terms included: elected (whether the chief was elected
into position) interacted with number of ruling families, elected with frequency of visits by
higher officials, elected with reaction of villagers when the chief misbehaves and economic
position of the chief interacted with villagers belief regarding the chiefs entitlement to
confiscate tax. The rationale for including these interaction effects is as follows. We expect
that elected chiefs may face a higher pressure to behave non-corruptly in villages with a
higher number of ruling families. The existence of more families who can provide a chief is a
credible threat to the chief’s power/position in case of misconduct. In a related argument,
elected chiefs are expected to be less corrupt if they are more frequently visited by higher
officials and/or more villagers react when the chief misbehaves. Within the context of our
theoretical framework, these two factors relate to the constraining effect of supervision and
monitoring on corruption behaviour by raising its expected cost. We included fixed effects
and community controls in this estimation. We again find a positive & significant effect (at 10
percent) of being born in the same village on the corruption of the chief. In addition, we again
find that chiefs are corrupt if the district commissioner is perceived to be corrupt (at 5
49
percent). We still do not find significant effects of the other variables and interaction terms on
corruption behaviour. In our last exercise, we changed our proxy for the corruption of the
chief & instead used the perception of the villagers regarding the corruption of the chief as
our dependent variable. This is indicated in column 2 of Table 4.2.2. We now find significant
effects of household size (negative), chief’s term in office (positive), protestant
religion(negative) and corruption of district commissioner(positive) on the chiefs’ corruption
behaviour ( all at 10 percent). It may be the case that chiefs with a larger household size may
have more income earning members and hence may not need to ‘steal’ community resources.
Or, such chiefs may be the richer chiefs who may have less incentive to divert community
resources (due to lower marginal benefit of corruption and/or higher marginal cost due to
social sanctions).
Table 4.2.2 Determinants of Corruption: Interaction effects
---------------------------------------------------------------Model 5
Variable
(D.V = Missing input)
Model 6
(D.V= Corruption perception)
------------------------------------------------------------------------------------------------
Age
Gender
ethorigin
Education
Eco_position
HHsize
Landsize
term
protestant
born
-0.046
-0.006
(0.07)
(0.04)
2.133
-2.271
(2.38)
(1.49)
0.355
-0.917
(1.73)
(1.26)
-0.267
-0.049
(0.20)
(0.11)
-0.336
-0.248
(5.13)
(0.75)
-0.061
-0.294*
(0.06)
(0.12)
0.001
0.007
(0.01)
(0.00)
-0.098
0.167*
(0.11)
(0.08)
0.060
-3.880*
(2.18)
(1.81)
3.935*
1.393
(1.61)
(0.85)
offfarm
2.659
0.346
(1.51)
(0.97)
elected
-9.281
-1.775
(12.49)
(1.01)
5.616
-0.065
(6.70)
(3.50)
shareliterate
50
rulingfamilies
ethnichetero
mobilephones
visits
corr_perception
0.189
-0.159
(0.44)
(0.15)
-1.942
1.376
(3.54)
(2.08)
0.002
-0.003
(0.00)
(0.00)
0.149
-3.485
(5.99)
(2.55)
1.257
(1.59)
Distance
-0.018
-0.065
(0.06)
(0.04)
Criticisms
-3.965
-5.183
(5.21)
(3.21)
Reaction
-0.190
0.215
positivedef
aventitlement
Corr-distr
elerul
(0.43)
(0.20)
-2.089
0.435
(2.60)
(1.47)
0.921
0.853
(3.38)
(0.66)
5.288**
3.178*
(1.92)
(1.29)
-0.476
(0.56)
elecvis
4.405
(8.60)
elecreact
0.789
econptax
0.283
(0.55)
(2.09)
constant
1.186
17.078
(19.24)
(11.32)
District FE
Yes
Yes
Community controls
Yes
Yes
--------------------------------------------------------------------------------------sigma
constant
5.385***
(0.51)
------------------------------------------------------------------------------------Chi2
54.995
68.422
PseudoR^2
0.109
0.484
N
111
111
------------------------------------------------------------------------------------*,** & *** represent significance at 10%, 5% & 1% respectively.
Standard errors in parenthesis
elerul= elected x rulingfamilies
elecvis= elected x visits
elecreact= elected x reaction
econptax= Eco_position X aventitlement
D.V= dependent variable
51
Some of our explanatory variables may be endogenous to corruption behaviour. But, as
explained in the previous chapter, we argue that the chiefs’ characteristics are exogenous and
the causation is most likely to go one way. An underlying assumption that we make here is
that chiefs who ‘steal’ the rice seeds are also most likely to (mis) use their position to
appropriate community resources for their private benefit.
52
Chapter Five: Conclusions & Reflections
5.1 Conclusions
In this study we tried to identify the micro-determinants of corruption of village chiefs in rural
Liberia. We employed household and community level data collected from 136 villages. We
used an objective measure to proxy corruption behaviour of the chiefs. We hypothesized that
corruption behaviour of chiefs is influenced by their socio-economic characteristics as well as
of their villages. We specified and estimated censored regression models to test our
hypotheses. We included district fixed effects to account for unobservable factors that may
vary at district level. We also included community controls to account for socio-economic
differences between the villages.
We find significant effects of being born in the same village and (perceived)corruption of
higher officials on the corruption behaviour of the chiefs. Chiefs who were born in the same
village are found to be corrupt as compared to their counter parts. Chiefs under a corrupt
district official are also found to be corrupt as compared to their counter parts. We explain
both of these findings in the context of our theoretical framework. Being born in same village
may be expected to reduce the marginal cost of engaging in corruption by reducing social
sanctions and supervision. Or, it may be expected to increase the marginal benefit of engaging
in corruption by raising the pressure to satisfy the needs a larger social network. To explain
our result regarding the effect of the (perceived) corruption of higher officials on the
corruption of the chief, we employ the insights from the Social learning theory. The chiefs
imitate the behaviour set by the district commissioner as it may be considered normal or
approved. This indicates the influence leaders have on their subordinates and suggests the
importance of tackling corruption at all levels and setting good leadership models for lower
level leaders. We do not find significant effects of the other variables included in our study on
corruption behaviour of the chiefs.
5.2 Critical Reflections
Though corruption has been a problem since time immemorial, it is still one of the most
difficult concepts to study. This is in part due to the difficulty of objectively measuring its
occurrence and the various forms in which it manifests itself. We tried to use a more objective
way of measuring corruption by accounting for missing rice seeds stored at chiefs’ house.
53
Even though such a measure may be regarded as objective, we cannot be sure if such a setting
replicates corruption environment in a real world. Since corruption is illegal or immoral in
most societies, individuals mostly engage in it if the likelihood of detection is zero or at least
negligible. But in our case, it seems most likely that the chiefs expect a measure of the rice
seeds allocated to the villages would be done before delivery and also expect the seeds to be
measured before they are distributed to the villagers. Hence, it may be the case that chiefs
anticipate higher likelihood of detection and refrain from ‘stealing’ the rice seeds. If some of
the chiefs still decide to steal part of the rice seeds in such a situation, this may also suggest
that such an exercise may be measuring their entitlement to a tax rather that corruption
behaviour. Looking at the large average landholding of the chiefs, it may also be possible that
we are measuring elite capture. Besides, there is no information on what happened to the
missing rice seeds(see Beekman & Bulte, 2015 for plausible alternative explanations). In fact,
there is low correlation (r= 0.2157) between our measure of corruption and the perceived
corruption of the chief by villagers.
We also believe that the insignificant results of some of the explanatory variables may be
related to the way we measured the variables. For example, we used the number of meals the
chief’s household had as indicator of economic status. But, this is just a crude measure. We do
not have data on the income of the chiefs from various sources ( even if they do not get salary
for their positions). We also do not have data on their assets, characteristics of their house,
expenditures and hence are not able to construct their socio-economic status index. In
addition, some of our variables may be endogenous to corruption behaviour. For example,
higher frequency of visits by higher officials may be expected to reduce corruption. But, on
the other hand, more corrupt chiefs may be more frequently visited by higher officials. In such
instances, the use of instrumental variables could have given us more insights but we are
unable to find suitable instruments in our case. Further research might give more insights on
the micro-determinants of corruption in rural Africa and establish causality by tackling these
limitations. It may be useful to combine lab-based measures with field experiments to have a
better measure corruption behaviour at micro-level and cross check the results of diffrent
approaches.
54
References
Abbink, K. and Serra, D. (2012). Anticorruption Policies: Lessons from the Lab. In: Serra, D.,
Wantchekon, L. (Eds.), New Advances in Experimental Research on Corruption.
Research in Experimental Economics, vol. 15. Emerald, Bingley, pp. 77-116.
Abbink, K., Arora, P., Gangadharan, L. and Wu, K. (2014). The Market for Corruption –
Evidence from a Natural Field Experiment .Forthcoming 2014.
Acemoglu, D. and Verdier, T. (1998). Property Rights, Corruption and the Allocation of
Talent:
a
General
Equilibrium
Approach.
The
Economic
Journal,108
(September),1381-1403.
Acemoglu, D., Reed, T. & Robinson, J. (2014). Chiefs: Economic Development and Elite
Control of Civil Society in Sierra Leone, Journal of Political Economy, 122(2), 319–
368.
Afrobarometer (2013). Summary of Results: Survey Overview The Quality of Democracy and
Governance in Liberia Afrobarometer Round 5, 2012.
Aidt, T. S. (2003), ‘Economic Analysis of Corruption: A Survey’, The Economic Journal.
113, F632–52.
Aidt, T. S. (2009) Corruption, institutions, and economic development, Oxford Review of
Economic Policy, 25(2), pp. 271–291. doi:10.1093/oxrep/grp012.
Aidt, T.S. (2011). The Causes of Corruption. A report for the Institute for Quantitative Social
Science at Harvard University.
Akers, R.L., Krohn, M. D., Lanza-Kaduze, L. and Radosevic, M. (1979). “Social Learning
and Deviant Behaviour: A Specific Test of A General Theory.” American Sociological
Review 44 (4): 636-655.
Andvig, J. C., and Moene, K. O. (1990), ‘How Corruption May Corrupt’, Journal of
Economic Behaviour and Organization, 13(1), 63–76.
55
Armantier, O., & Boly, A. (2011). A Controlled Field Experiment on Corruption. European
Economic Review, 55(8), 1072-1082.
Armantier, O., and Boly, A. (2012). On the External Validity of Laboratory Experiments on
Corruption. In: Serra, D., Wantchekon, L. (Eds.), New Advances in Experimental
Research on Corruption. Research in Experimental Economics, vol. 15. Emerald,
Bingley, pp. 117-144.
Baldwin, K. & Mvukiyehe, E. (2015) Elections and collective action: Evidence from Changes
in Traditional Institutions in Liberia, World Politics (forthcoming).
Barr, A. and Serra, D. (2009). The Effect of Externalities and Framing on Bribery in a Petty
Corruption Experiment. Experimental Economics, 12 (4), 488–503.
Barr, A. and Serra, D. (2010). Corruption and Culture: An experimental analysis. Journal of
Public Economics.94, 862–869.
Bates, R. (1981). States and Markets in Tropical Africa: The Political Basis of Agricultural
Policy. Berkeley: University of California Press.
Beekman, G., Bulte, E., Nillesen, E. (2013). Corruption and Economic Activity: Micro Level
Evidence from Rural Liberia. European Journal of Political Economy. 30, 70–79.
Beekman, G., Bulte, E., Nillesen, E. (2014). Corruption, Investment and Contributions to
Public Goods: Experimental Evidence from Rural Liberia. Journal of Public
Economics.115,37-47.
Beekman, G. and Bulte, E. (2015). A Note on Targeting by Predatory Leaders: Evidence from
Rural Liberia, Oxford Development Studies.
Beekman, G. , Bulte, E., Kusters, C., Peters, B. and Trijsburg, S. (2015). Joint MFS-II
Evaluation Liberia: Synthesis Report. Wageningen University and Research Centre.
56
Boucher, A.J., Durch, W.J., Midyette, M., Rose, S. and Jason Terry, J. (2007). Mapping and
Fighting Corruption in War-torn States. Stimson Center Report NO. 61. The Henry L.
Stimson Center, Washington, DC.
Campos, E.J., Lien, D. and Pradhan, S. (1999) The Impact of Corruption on Investment:
Predictability Matters. World Development 27: 1059-1067.
Chaudhuri, A. (2012). Gender and Corruption: A Survey of the Experimental Evidence. In:
Serra, D., Wantchekon, L. (Eds.), New Advances in Experimental Research on
Corruption. Research in Experimental Economics, vol. 15. Emerald, Bingley, pp. 1350.
Cheng, C. and Zaum, D. (eds), (2011). Corruption and Post-Conflict Peacebuilding; selling
the Peace? Eds. Routledhe, Taylor and Francis Group, London and New York.
Clark, M. (2008). Combating corruption in Liberia: assessing the impact of the Governance
and Economic Management Assistance program (GEMAP)
Donovan, J.E., Jessor, R., and Costa, F.M. (1991). Adolescent Health Behaviour and
Conventionality-Unconventionality: An Extension of Problem Behaviour Theory.
Health Psychology. 10(1), 52-61.
Economist (2015, August 27). Ebola in Graphics: The toll of a tragedy. Published online on
27
August,
2015.
Retrieved
from
http://www.economist.com/blogs/graphicdetail/2015/08/ebola-graphics
Egger. P and H. Winner. (2005). Evidence on Corruption as an Incentive for Foreign Direct
Investment. European Journal of Political Economy, Vol. 21, pp.932-952.
Fearon, J.D. and Laitin, D.D. (1996). Explaining Interethnic Cooperation. The American
Political Science Review. Vol. 90, No. 4, pp. 715-735
Graaf, G. (2007): Causes of Corruption: Towards a Contextual Theory of Corruption. In:
Public Administration Quarterly, 31, 1, pp. 39-86.
57
Humphreys, M., Serra, R. and Windt, P. (2013). Exporting Institutions: Evidence from a
Field Experiment in Eastern Congo, Working Paper. New York: Columbia University.
Jones, B. and Olken, B. (2005). Do Leaders Matter? National Leadership and Growth Since
World War II. Quarterly Journal of Economics 120: 835-864.
Lambsdorff, J.G. (2005). Causes and Consequences of Corruption - What do we know from a
cross-section
of
countries?
Volkswirtschaftliche
Reihe
ISSN
1435-3520,
Diskussionsbeitrag Nr. V-34-05.
Mauro, P. (1995). Corruption and Growth. Quarterly Journal of Economics 110: 681-712.
Meon, P.G. and Sekkat, K. (2005). Does corruption grease or sand the wheels of growth?
Journal of Public Choice. 122: 69–97.
Olken, B. (2006). Corruption and the Costs of Redistribution: Micro evidence from Indonesia.
Journal of Public Economics. 90: 853-870.
Olken, B. (2007). Monitoring Corruption: Evidence from a Field Experiment in Indonesia.
Journal of Political Economy. 115, 200–249.
Olken, B. and Pande, R. (2012). Corruption in Developing Countries. The Annual Review of
Economics. 4, 479–509.
Reinikka R. and J. Svensson (2004). Local Capture: Evidence from a Central Government
Transfer Program in Uganda. Quarterly Journal of Economics 119: 679-706.
Richards, P., S. Archibald, B. Bruce, W. Modad, E. Mulbah, T. Varpilah, and J. Vincent
(2005). Community Cohesion in Liberia: A Post-War Rapid Social Assessment. Social
Development Papers: Conflict Prevention & Reconstruction Paper # 21. Washington
DC: World Bank.
Rose-Ackerman, S. (2008). Corruption. In: Rowley, C.K. and
Schneider, F.G. (eds.),
Readings in Public Choice and Constitutional Political Economy. Springer Science
and Business Media. PP. 551-566.
58
Schleifer, A. and R. Vishny (1993). Corruption. Quarterly Journal of Economics 108: 207230.
Seldadyo and Haan (2006). The Determinants of Corruption: A Literature Survey and New
Evidence. Paper Prepared for the 2006 EPCS Conference, Turku, Finland, 20-23 April
2006.
Sequeira, S. (2012). Advances in Measuring Corruption in the Field. In: Serra, D.,
Wantchekon, L. (Eds.), New Advances in Experimental Research on Corruption.
Research in Experimental Economics, vol. 15. Emerald, Bingley, pp. 145–176.
Serra, D. (2006). Empirical Determinants of Corruption: A Sensitivity Analysis. Journal of
Public Choice 12. 225-256.
Serra, D. (2011). Combining Top-down and Bottom up Accountability: Evidence from
Bribery
Experiment.
Journal
of
Law,
Economics
and
Organization.
Doi:10.1093/jleo/ewr010.
Serra, D. and Wantchekon, L. (2012). Experimental Research on Corruption: Introduction and
Overview. In: Serra, D., Wantchekon, L. (Eds.), New Advances in Experimental
Research on Corruption. Research in Experimental Economics, vol. 15. Emerald,
Bingley, pp. 1-11.
Serra, D. & Wantchekon, L. (Eds.), (2012). New Advances in Experimental Research on
Corruption. Research in Experimental Economics, volume 15. Emerald, Bingley UK.
Soreide, T. (2014). Drivers of Corruption: A Brief review. World Bank Studies. Washington,
DC: World Bank. doi: 10.1596/978-1-4648-0401-4. License: Creative Commons
Attribution CC By3.0 IGO.
Svensson, J. (2003). Who Must Pay Bribes and How Much? Quarterly Journal of Economics.
118, 207–230.
59
Tavits, M. (2005). Causes of corruption: Testing competing hypotheses. In Joint Sessions of
Workshops, European Consortium for Political Research, Nicosia, Cyprus, April,
2005.
Transparency International (2012). Overview of Corruption and anti-corruption in Liberia.
Anti-corruption
Resource
Centre.U4
Expert
Answer.
Retrieved
from
http://www.u4.no/publications/overview-of-corruption-and-anti-corruption-in-liberia/
Transparency International (2015). What are the Costs of Corruption? Retrieved from
https://www.transparency.org/what-is-corruption#define
Treisman, D. (2000). ”The Causes of Corruption: A Cross-National Study”. Journal of Public
Economics 76: 399-457.
Treisman, D. (2007). What have We Learned about the Causes of Corruption from Ten Years
of Cross-national Empirical Research? Annu. Rev. Polit. Sci. 2007. 10:211–44.
Turley, T., Voors, M., Bulte, E., Kontoleon, A., and List, J. (2013). Chiefs for a Day: Elite
Capture and Management Performance. Evidence from a Field Experiment in Sierra
Leone.
UNDP (2014). Human Development Report 2014. Sustaining Human Progress: Reducing
Vulnerabilities and Building Resilience. PBM Graphics, New York, NY.
US department of State (2010). Report on human right practices 2010. Retrieved from
http://www.state.gov/documents/organization/160129.pdf
Winters, M.S., Testa, P., Fredrickson, M.M. (2012). Using Field Experiments to Understand
Information as an Antidote to Corruption. In: Serra, D., Wantchekon, L. (Eds.), New
Advances in Experimental Research on Corruption. Research in Experimental
Economics, vol. 15. Emerald, Bingley, pp. 213-246.
World Bank (2013). Governance and Public Sector Management. Retrieved from
http://web.worldbank.org/WBSITE/EXTERNAL/TOPICS/EXTPUBLICSECTORAN
60
DGOVERNANCE/0,,contentMDK:23493111~pagePK:148956~piPK:216618~theSite
PK:286305,00.html
World Bank (2015). Country Data. Retrieved from http://data.worldbank.org/country/liberia
61